SPAD ranging method and system based on dynamic histogram storage and multi-frame fusion
The SPAD ranging method, which uses multi-frame delay adjustment and dynamic histogram storage, solves the problem of balancing ranging accuracy, storage resources, and power consumption in existing technologies, and achieves high ranging performance and flexible frame rate adjustment under limited hardware resources.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- XIDIAN UNIV
- Filing Date
- 2025-12-29
- Publication Date
- 2026-05-08
AI Technical Summary
Existing SPAD ranging technology suffers from a tradeoff between ranging accuracy, on-chip storage resources, power consumption, and measurement frame rate, with performance dropping sharply, especially in environments with low signal-to-noise ratio or strong background light.
The single-frame measurement period is divided into multiple subframes by a multi-frame delay adjustment step. By combining photon filtering and dynamic histogram storage methods with multi-frame fusion technology, extreme value suppression of noise and sparsification accumulation of correlation signals are achieved. Finally, high-precision time of flight is extracted through nonlinear fitting.
With limited hardware resources, it significantly reduces storage overhead and power consumption, while supporting flexible switching between ranging accuracy and frame rate, achieving a balance between efficient collaboration and optimization to meet the needs of different application scenarios.
Smart Images
Figure CN121995387A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of single-photon detection and integrated circuit technology, specifically relating to a SPAD ranging method and system based on dynamic histogram storage and multi-frame fusion. Background Technology
[0002] Single-photon avalanche diodes (SPADs), with their single-photon-level detection sensitivity and picosecond-level time resolution, have become core detectors in cutting-edge fields such as lidar, 3D sensing, and low-light imaging. SPAD-based ranging systems typically employ the time-of-flight (ToF) principle, calculating distance by measuring the time delay between the emitted laser wave and the echo photon. Time-correlated single-photon counting (TCSPC) is the mainstream technology for achieving high-precision ranging. This method requires high-resolution quantization of the arrival time of the echo photons within repeated laser cycles, statistical analysis of their time distribution to construct a histogram, and finally, extraction of the precise time of flight by identifying the peak positions of the histogram.
[0003] However, to achieve full-range, high-precision measurement, traditional TCSPC solutions require a high-precision time-to-digital converter (TDC) and large-capacity on-chip memory within each pixel or macropixel to cover the entire time window and record a complete time histogram. This results in enormous hardware resource consumption, significantly increasing chip area, power consumption, and cost, severely restricting the development of SPAD arrays towards large-scale, highly integrated designs. To reduce resource overhead, several improvement paths exist in existing technologies: one is to reduce storage overhead by lowering the time resolution, compressing the histogram dimension, or shortening the measurement window, but this method usually sacrifices the system's ranging accuracy and dynamic range, especially with a sharp performance drop in low signal-to-noise ratio or strong background light environments; the other is to improve accuracy by using multi-frame accumulation or sub-frame phase scanning, but this requires storing and processing multiple frames of complete histograms, which not only further increases storage and read / write power consumption but also leads to a decrease in system frame rate, making it difficult to meet the real-time requirements of high-dynamic scenes.
[0004] Therefore, existing SPAD ranging technology still faces a dilemma in balancing ranging accuracy, on-chip storage resources, power consumption, and measurement frame rate. Summary of the Invention
[0005] To address the aforementioned problems in the existing technology, this invention provides a SPAD ranging method and system based on dynamic histogram storage and multi-frame fusion. The technical problem to be solved by this invention is achieved through the following technical solution: This invention provides a SPAD ranging method and system based on dynamic histogram storage and multi-frame fusion, the method comprising: Multi-frame delay adjustment steps: Divide the single-frame measurement period into multiple sub-frames, and configure a corresponding phase delay amount for each sub-frame so that each sub-frame corresponds to a different equivalent sampling time reference; wherein, each sub-frame corresponds to multiple exposure periods; For each subframe, the photon filtering step and the dynamic histogram storage step are performed sequentially, where: The photon filtering step includes: within each exposure cycle, adaptively filtering photon triggering events of multiple SPADs in macropixels by extreme value tracking, and outputting the effective photon response time and corresponding time calibration information and trigger intensity; The dynamic histogram storage step includes: within each exposure cycle, based on the output of the photon screening step in the current exposure cycle, dynamically tracking the return time of the echo photons and generating the corresponding processing result; integrating the processing results of all exposure cycles in the current sub-frame to generate the candidate histogram corresponding to the current sub-frame; Fitting process steps: Based on the candidate histograms corresponding to all subframes and their respective phase delays, a fused histogram with a higher temporal resolution than that of a single subframe is reconstructed through data fusion. Using the time-of-flight extraction algorithm, waveform analysis and nonlinear fitting are performed on the fused histogram to extract the flight time of the echo photons and output the ranging results.
[0006] Compared with the prior art, the beneficial effects of the present invention are as follows: To address the challenge of balancing ranging accuracy, on-chip storage resources, power consumption, and measurement frame rate in existing SPAD ranging technologies, this invention provides a SPAD ranging method and system based on dynamic histogram storage and multi-frame fusion. This method establishes a phase-separated sampling benchmark by introducing multi-frame delay adjustment, performs extreme noise suppression using photon filtering, and utilizes dynamic histogram storage to achieve sparse accumulation only for relevant signal events. Finally, high-precision time-of-flight is extracted through multi-frame data fusion and nonlinear fitting. This method requires only the maintenance of a small amount of valid event information in hardware, significantly reducing storage overhead and power consumption. It also supports flexible switching between measurement accuracy and frame rate through a configurable number of subframes, thereby achieving efficient coordination and optimized balance between ranging accuracy, on-chip resources, power consumption, and frame rate at the system level. This provides an effective technical path for achieving high-resolution 3D imaging under limited hardware resources. Attached Figure Description
[0007] Figure 1 This is a schematic diagram of the data flow of the SPAD ranging method based on dynamic histogram storage and multi-frame fusion provided in the embodiments of the present invention; Figure 2 This is a schematic diagram of the operation of the SPAD ranging system based on dynamic histogram storage and multi-frame fusion provided in the embodiments of the present invention. Detailed Implementation
[0008] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.
[0009] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. In addition, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.
[0010] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings, disclosure, and appended claims in carrying out the claimed invention. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.
[0011] The SPAD ranging method and system based on dynamic histogram storage and multi-frame fusion proposed in this invention will now be described in detail with reference to the accompanying drawings.
[0012] Figure 1 This is a schematic diagram of the data flow of the SPAD ranging method based on dynamic histogram storage and multi-frame fusion provided in an embodiment of the present invention. For example... Figure 1 As shown, the method includes steps 110-140. Specifically: S110: Multi-frame delay adjustment steps: Divide the single-frame measurement period into multiple sub-frames and configure a corresponding phase delay amount for each sub-frame so that each sub-frame corresponds to a different equivalent sampling time reference; wherein, each sub-frame corresponds to multiple exposure periods.
[0013] Here, the goal of S110 is to improve the overall temporal resolution of the system through equivalent sampling without increasing the hardware cost of a single TDC resolution. Specifically, based on the ranging requirements, a single frame measurement period is divided into multiple subframes, and a delay selection signal corresponding to each subframe is generated; multiple delay signals with different fixed delay amounts are generated using an open-loop delay chain; based on the delay selection signal corresponding to a single subframe, one of the multiple delay signals is selected, and its corresponding sampling clock is adjusted so that the sampling time of that subframe is offset from the laser emission time by the corresponding phase delay amount, thereby establishing an equivalent sampling time reference for that subframe.
[0014] It should be understood that the temporal resolution of a single-frame measurement is limited by the hardware circuitry. Dividing the single-frame measurement period into multiple subframes is to improve temporal resolution without increasing hardware complexity. If only a single frame is used, the temporal resolution is limited by the hardware's sampling interval, such as the minimum resolution unit of a TDC. By using multiple subframes, each introducing a different phase delay, it's equivalent to a sub-pixel-level sampling offset on the time axis. Then, a delay selection signal is generated to precisely control the amount of phase delay for each subframe. The delay chain generates a series of fixed-delay delay signals, and the delay selection signal ensures that each subframe uses a different delay signal, thus covering a denser sampling point on the time axis. In this way, the measurement results of multiple subframes are staggered on the time axis, but the delay amounts are known, allowing for subsequent data fusion to reconstruct a higher-resolution histogram.
[0015] Here, the method also includes a working mode switching step: in response to the accuracy priority instruction, the number of subframes in a single frame measurement period is increased; in response to the frame rate priority instruction, the number of subframes in a single frame measurement period is reduced, or the flight time is directly extracted based on the candidate histogram of a single subframe to output a high frame rate ranging result.
[0016] The reason lies in the inherent trade-off between accuracy and frame rate in traditional SPAD ranging. To address this inherent contradiction, the system dynamically configures the number of subframes and processing flow, allowing it to adaptively switch between two optimal operating points based on actual scenario requirements. In accuracy-first mode, the number of subframes is increased, and a complete fusion fitting process is executed, sacrificing a lower frame rate for extremely high temporal resolution and ranging accuracy. In frame rate-first mode, the number of subframes is reduced, or even fitting is skipped, resulting in direct output. This sacrifices accuracy for high response speed and real-time measurement. In this way, the same hardware system can flexibly adapt to different application scenarios, from high-dynamic rapid sensing to high-precision fine reconstruction, achieving a dynamically optimal configuration between system resources and performance.
[0017] Here, the open-loop delay chain consists of multiple cascaded CMOS inverters. Each CMOS inverter has the same structure. By adjusting the size of the CMOS inverters and the load capacitance, the delay of each delay unit can be stabilized at a predetermined fixed value, such as Δt. Assuming the open-loop delay chain generates N delayed signals, the delay of each signal is 1×Δt, 2×Δt, ..., N×Δt, respectively. To ensure consistent driving capability of each delayed signal and to avoid signal attenuation, buffers are used to process each delayed signal.
[0018] Here, a delay selection circuit is used to generate the delay selection signal. The delay selection circuit employs a multiplexer architecture, selecting one output from N delayed signals based on the system's operating mode and the current subframe's state to drive the photon filtering step. Different subframes correspond to different delay selection signals, ensuring that each subframe's time sampling has a preset delay offset. For example, when the system has 10 subframes and the open-loop delay chain contains 10 delay units, each subframe sequentially selects a delay signal from 1×Δt to 10×Δt, enabling the 10 subframes to form step-by-step time sampling within a single frame period, achieving an equivalent Δt-level time resolution.
[0019] After splitting the image into multiple subframes, steps S120 and S130 are executed sequentially in each exposure cycle until the calculation for all exposure cycles is completed. Specifically: For each subframe, the photon filtering step and the dynamic histogram storage step are performed sequentially, where: S120: The photon selection step includes: within each exposure cycle, adaptively selecting photon triggering events of multiple SPADs in macropixels through extreme value tracking, and outputting the effective photon response time and corresponding time calibration information and trigger intensity.
[0020] Specifically, the photon screening step includes: within the current exposure cycle, using extreme value tracking, counting the number of photon triggers of multiple SPADs in the macropixel within a summation window, and using this count as the trigger intensity at the current sampling moment; comparing the trigger intensity at the current sampling moment with the current maximum trigger intensity; if the trigger intensity at the current sampling moment is greater, updating the current maximum trigger intensity using the trigger intensity at the current sampling moment, and updating the current candidate response moment using the current sampling moment; where the current maximum trigger intensity is the maximum number of triggers at the previous sampling moment, and the current candidate response moment is the sampling moment corresponding to the current maximum trigger intensity; otherwise, not updating the current maximum trigger intensity and the current candidate response moment; after the current exposure cycle ends, outputting the current candidate response moment as the corresponding effective photon response moment; quantifying the time interval between the effective photon response moment and the laser emission moment as time calibration information.
[0021] Here, the summation window is a dynamic time window, whose width and / or position are dynamically determined based on the spatiotemporal distribution characteristics of photon-triggered events in the macro-pixel. Specifically, the more photon triggers, the more likely the triggering event is caused by a valid echo photon, and the higher the corresponding weighting factor is assigned. Through nonlinear weighting, the equivalent triggering probability of echo photon events can be increased in subsequent statistical processes, thereby enhancing the prominence of the echo signal in the histogram statistical results.
[0022] For example, suppose a macropixel consists of 16 SPADs in a 4×4 configuration, with a summation window (i.e., a statistical time window) set to 1 nanosecond (ns), a current exposure cycle of 100 ns, and a sampling clock cycle of 1 ns (i.e., intensity statistics and comparisons are performed every 1 ns). At a certain sampling moment (e.g., the 23rd ns), the system statistically analyzes the triggering status of the 16 SPADs over the past 1 ns (i.e., from the 22nd to the 23rd ns): SPADs 0, 3, 5, 7, and 12 each trigger once (due to background noise or dark count); SPADs 8, 9, 10, and 11 trigger once each due to the spatial diffusion effect of the same echo photon; other SPADs do not trigger. Therefore, the total number of triggers within this summation window is 8, and this value is recorded as the "trigger intensity at the current sampling moment (the 23rd ns)". Throughout the entire exposure cycle (100 ns), the above process is repeated every 1 ns. The system compares the trigger intensity at each sampling moment with the stored current maximum trigger intensity in real time. If the intensity at 23ns (8 times) is greater than the previously stored maximum value (e.g., 5 times at 15ns), then the current maximum trigger intensity is updated to 8 times, and the candidate response time is updated to 23ns. If the intensity at subsequent sampling times (e.g., 50ns) is 6 times, which does not exceed 8 times, then the current maximum value and time scale remain unchanged.
[0023] Through this extreme value tracking mechanism, the system ultimately outputs the moment with the maximum trigger intensity (8 times) (at the 23rd ns) after the end of the exposure cycle as the unique and most effective photon response moment, and simultaneously outputs its corresponding intensity value (8 times) as the trigger intensity. This design ensures that the system can still filter out the key moment with the most concentrated signal and most likely to represent the true echo under strong background light or noise interference, thereby significantly reducing the amount of data processed subsequently.
[0024] Here, the time calibration information is obtained through a time-to-digital converter (TDC). For example, once the effective photon response time is determined to be 23 nanoseconds (ns), the system initiates the TDC to quantize the time interval between this time and the synchronization start time of laser emission. Assume the TDC uses a delay chain-based counting structure with a time resolution of 10 picoseconds. If the laser emission synchronization time is 0 ns, the TDC will measure the precise delay from 0 ns to 23 ns. The quantization process is as follows: an integer number of clock cycles (e.g., 5 ns per clock cycle, corresponding to 4 full cycles totaling 20 ns) are recorded using a high-speed clock counter, and then the remaining tail time (e.g., 3.012 ns) is measured using a fine delay chain. Finally, the TDC outputs a precise digital timestamp, such as 23012 (unit: 10 ps), which serves as the time calibration information. This information, along with the effective photon response time (23 ns) and its trigger intensity (e.g., 8 times), is output to provide a high-precision time reference for time correlation analysis in the subsequent dynamic histogram storage step and multi-frame data alignment in the fitting process step.
[0025] Through TDC quantization, the system converts the analog arrival time of photons into high-precision time data that can be processed by digital circuits, which is the key to realizing sub-nanosecond or even picosecond time-of-flight measurement.
[0026] S130: The dynamic histogram storage step includes: within each exposure cycle, based on the output of the photon screening step in the current exposure cycle, dynamically tracking the return time of the echo photons and generating the corresponding processing result; integrating the processing results of all exposure cycles in the current subframe to generate the candidate histogram corresponding to the current subframe.
[0027] Here, the dynamic histogram storage step includes: nonlinearly weighting the number of photon triggers corresponding to the effective photon response times output by the photon filtering step within the current exposure cycle to obtain the weighted trigger intensity; based on the weighted trigger intensity and the corresponding time calibration information, analyzing the distribution correlation between the photon trigger events corresponding to the weighted trigger intensity and the preset event information in the time dimension, updating the trigger times of photon trigger events that meet the preset correlation conditions to candidate return times, and iteratively updating the event information in the dynamic histogram storage module; wherein, the preset event information refers to the information of photon trigger events that meet the correlation conditions stored at the end of the previous exposure cycle; after the current exposure cycle ends, acquiring multiple accumulated candidate return times, storing them sequentially in descending order of trigger intensity, and generating the corresponding candidate histograms; summarizing the candidate histograms corresponding to all exposure cycles to obtain the candidate histogram of the current subframe.
[0028] It should be noted that the dynamic histogram storage module stores the data output from the photon selection step within each exposure cycle. Its event information includes, but is not limited to, the photon trigger time and the number of triggers.
[0029] For example, suppose the current subframe contains 3 exposure cycles, and the photon filtering step outputs a set of data (effective photon response time, time calibration information, trigger intensity) in each exposure cycle, and then: In the first exposure cycle, the photon selection step outputs the effective photon response time T1 (corresponding to time calibration information 23.012ns), with 8 trigger intensities. The dynamic histogram storage step performs non-linear weighting on this (e.g., weighting coefficient 1.5), resulting in a weighted intensity of 12. Since the dynamic histogram storage module is empty at this time (no historical information), the system directly sets T1 and its time calibration information (23.012ns) as the candidate return time and stores them in the dynamic histogram storage module.
[0030] In the second exposure cycle, the photon filtering step outputs a valid photon response time T2 (corresponding to time calibration information of 23.125 ns), with 6 trigger intensities and a weighted intensity of 9. The system reads the event information stored in the previous cycle (including time calibration information of 23.012 ns) and calculates the time difference between the two based on the time calibration information output in the current cycle (23.125 ns), which is 0.113 ns. The system determines whether this difference is within a preset time correlation window (e.g., 0.5 ns). Since 0.113 ns < 0.5 ns, T2 and T1 are determined to be highly correlated in time. Therefore, T2 and its time calibration information (23.125 ns) are also updated as candidate return times, and the event set in the storage module is iteratively updated.
[0031] In the third exposure cycle, the output time T3 (corresponding to time calibration information 45.200ns) is calculated with 10 intensity increments, resulting in a weighted intensity of 15. The system again uses its time calibration information (45.200ns) to compare it one by one with the time calibration information of existing events in the storage module (23.012ns, 23.125ns). The calculated time difference is approximately 22ns, far exceeding the correlation window, and is therefore determined to be irrelevant, possibly originating from new noise or another target. Therefore, T3 is not adopted.
[0032] Finally, the processing results of the three exposure cycles are integrated within this subframe to generate candidate histograms sorted by weighted intensity: {time calibration information: 23012 (23.012ns), weighted intensity: 12; time calibration information: 23125 (23.125ns), weighted intensity: 9}.
[0033] In other words, S130 weights and strengthens the effective signal, eliminates discrete noise through correlation analysis, and iteratively updates the focus on the true echo. Ultimately, only a few highly correlated and prominent moments are stored in the candidate histogram, thus preserving key echo features with extremely low storage overhead.
[0034] S140: Fitting Processing Steps: Based on the candidate histograms corresponding to all subframes and their respective phase delays, a fused histogram with a higher temporal resolution than that of a single subframe is reconstructed through data fusion. Using a time-of-flight extraction algorithm, waveform analysis and nonlinear fitting are performed on the fused histogram to extract the flight time of the echo photons and output the ranging results.
[0035] Specifically, the fitting process includes: aligning and fusing the data in the corresponding candidate histograms on the time axis according to the phase delay of each subframe, and reconstructing a fused histogram with a time resolution higher than the measurement resolution of any subframe; analyzing the waveform distribution characteristics of the fused histogram and determining the final return time of the echo photon through nonlinear fitting; and calculating the final ranging result based on the time difference between the final return time and the laser emission time, combined with the speed of light.
[0036] For example, assume the system divides a single-frame measurement period into three subframes (subframes A, B, and C), and configures different phase delays for each subframe (ΔA=0ps, ΔB=33ps, and ΔC=67ps). The candidate histogram of subframe A (ΔA=0ps) is recorded to the following main feature points: time calibration information 10000 (corresponding to 10.000ns), weighting intensity 15; the candidate histogram of subframe B (ΔB=33ps) is recorded to the following main feature points: time calibration information 9967 (corresponding to 9.967ns), weighting intensity 12; the candidate histogram of subframe C (ΔC=67ps) is recorded to the following main feature points: time calibration information 9933 (corresponding to 9.933ns), weighting intensity 10.
[0037] First, the known phase delays (ΔA, ΔB, and ΔC) for each subframe are obtained. Then, the data in each candidate histogram is compensated for to remove the subframe-specific delays and aligned to a unified absolute time base (laser emission time). For example: The data in subframe A remains unchanged (because ΔA=0), and the absolute time point is still 10.000ns; Subframe B data compensation +33ps, absolute time point: 9.967ns + 0.033ns = 10.000ns; Subframe C data compensation +67ps, absolute time point is 9.933ns + 0.067ns = 10.000ns; At this point, it can be seen that the feature points with the highest intensity in the three subframes converge at the same absolute time point (10.000 ns) after the delay is compensated. The system fuses these aligned data points (e.g., by intensity superposition or weighted averaging) to generate a new fused histogram with denser data points. Since the fusion process makes comprehensive use of the information from the "misaligned" sampling (interval of about 33 ps) of the three subframes on the original time axis, the equivalent temporal resolution of this fused histogram (~33 ps) is much higher than the original measurement resolution of any single subframe (assumed to be 100 ps).
[0038] Subsequently, the waveform distribution of the fused histogram around 10.000 ns is analyzed. Due to the data originating from different phases and undergoing intensity weighting, this location forms a peak with a high signal-to-noise ratio and distinct characteristics. The module employs a nonlinear fitting algorithm (such as Gaussian fitting) to model the peak and its neighboring data points, thereby determining the center position of the peak with sub-pixel accuracy. For example, the fitting result might determine the final return time as 10.002 ns.
[0039] Finally, based on the final return time of 10.002 ns obtained from the fitting, the time difference between it and the laser emission time, i.e., the time of flight (ToF), is calculated to be 10.002 ns. Using the speed of light (approximately 0.299792458 m / ns), the target distance is calculated to be approximately 1.500 meters (calculation: 10.002 ns × 0.299792458 m / ns ≈ 2.999 meters, round trip distance divided by 2 gives approximately 1.500 meters).
[0040] Corresponding to the SPAD ranging method based on dynamic histogram storage and multi-frame fusion provided in the embodiments of the present invention, the embodiments of the present invention also provide a SPAD ranging system based on dynamic histogram storage and multi-frame fusion. The system deploys the aforementioned SPAD ranging method; the system includes: a multi-frame delay adjustment module, a photon filtering module, a dynamic histogram storage module, and a fitting processing module; wherein: The multi-frame delay adjustment module is used to divide a single-frame measurement cycle into multiple sub-frames and configure a corresponding phase delay amount for each sub-frame so that each sub-frame corresponds to a different equivalent sampling time reference; wherein, each sub-frame corresponds to multiple exposure cycles; The photon filtering module and the dynamic histogram storage module process each subframe; where: The photon filtering module is used to adaptively filter photon triggering events of multiple SPADs in macropixels in each exposure cycle by extreme value tracking, suppress invalid triggering caused by background noise, and output the effective photon response time and corresponding time calibration information and trigger intensity. The dynamic histogram storage module is used to dynamically track the return time of echo photons based on the output of the photon screening step in the current exposure cycle, and generate the corresponding processing results in each exposure cycle; it integrates the processing results of all exposure cycles in the current subframe to generate the candidate histogram corresponding to the current subframe. The fitting processing module is used to reconstruct a fused histogram with a higher temporal resolution than that of a single subframe based on the candidate histograms corresponding to all subframes and their respective phase delays. Using the time-of-flight extraction algorithm, the fused histogram is subjected to waveform analysis and nonlinear fitting to extract the flight time of the echo photons and output the ranging results.
[0041] Figure 2 This is a schematic diagram of the operation of the SPAD ranging system based on dynamic histogram storage and multi-frame fusion provided in an embodiment of the present invention. Now, in conjunction with... Figure 2 Describe the specific architecture of each module. For example... Figure 2 As shown, Assuming the system needs to measure the distance to a stationary target 1.5 meters away, the single-frame measurement period is divided into 3 subframes (N=3), with M=100 exposures performed in each subframe. Each module is based on... Figure 2 The architecture works collaboratively, and its internal components and data flow are as follows: 1) Multi-frame delay adjustment module (corresponding to: state control circuit, delay chain circuit, and delay selection circuit). Among them, State control circuit: Based on the configuration, it generates the three subframe numbers (#1, #2, #3) of the current frame and the corresponding delay selection signals (such as 00, 01, 10). Delay chain circuit: It is composed of multiple cascaded CMOS inverters (existing mature circuit). After inputting a reference clock, it generates multiple signals with fixed delay steps (such as Δt=33ps). Delay selection circuit: According to the instructions of the state control module, it selects different delay signals for the three subframes in sequence.
[0042] 2) Photon filtering module (including: summation circuit, comparison circuit, photon event candidate circuit, and synchronous sampling circuit); this module operates independently in each exposure cycle. Taking the first exposure cycle of subframe #1 as an example: Summation circuit (existing circuit): Within a preset dynamic summation window, the number of triggers of all SPADs in the macro pixel (such as a 4x4 SPAD array) is counted in real time; Comparison circuit and photon event candidate circuit (existing combinational logic and registers): During the current exposure cycle, continuously compare the current summation value with the maximum value stored in the cycle; if the current value is larger, update the candidate maximum value and the corresponding sampling time; at the end of the current exposure cycle, the circuit locks and outputs the effective photon response time (a time point) and the corresponding trigger intensity (count value); Synchronous sampling circuit (TDC, existing high-precision time measurement circuit): Receives the laser synchronization signal and the determined response time, quantizes the time interval between the two, and outputs high-precision time calibration information (digital timestamp).
[0043] 3) Dynamic histogram storage module (including intensity weighting circuit, iteration circuit, sorting circuit, and storage circuit). This module also receives and processes the output of the photon filtering module in each exposure cycle. Intensity weighting circuit (existing digital multiplication / lookup table circuit): Nonlinearly weights the input trigger intensity (e.g., the higher the intensity, the larger the weighting coefficient) to obtain the weighted trigger intensity; Iterative circuit (existing state machine and correlator logic): Based on the time calibration information and weighted trigger intensity of the current exposure cycle, as well as the event information (timestamp and intensity) of the previous cycle or historical accumulation read from the storage circuit, it performs correlation analysis in the time dimension (such as determining whether it is within a preset time window); only events that pass the correlation test are determined as valid candidate return times; Storage and sorting circuits (existing on-chip SRAM / registers and sorting logic): The iterative circuit writes the valid candidate return times and their weighted intensities into the storage circuit; after all exposure cycles of a subframe are completed, the sorting circuit sorts all events accumulated in the storage circuit for that subframe in descending order of weighted intensities, and finally outputs the candidate histogram (a sparse, ordered time-intensity list) corresponding to that subframe.
[0044] 4) Fitting Processing Module (corresponding software / firmware units: reconstructed histogram distribution feature extraction algorithm, time-of-flight extraction algorithm). This module receives candidate histograms of all subframes (#1, #2, #3) and the phase delay amounts (Δ1, Δ2, Δ3) of each subframe from the multi-frame delay adjustment module: The algorithm for reconstructing histogram distribution features is as follows: First, the phase delay of each subframe is used to compensate for all timestamps in the candidate histogram (i.e., the corresponding delay value is subtracted) to align all data onto a unified absolute time axis; then, the aligned data is interpolated and fused to generate a fused histogram with denser data points and an equivalent time resolution (~33ps) higher than the original measurement resolution of any single subframe (e.g., 100ps). Time-of-flight extraction algorithm: Peak detection and nonlinear fitting (such as Gaussian fitting) are performed on the waveform of the fused histogram to determine the final return time of the echo signal with sub-pixel accuracy; finally, the time of flight is calculated based on the difference between this time and the laser emission time, and the distance value (e.g., 1.500 meters) is converted and output.
[0045] It should be understood that the specific operation process can be referred to each step in the SPAD ranging method based on dynamic histogram storage and multi-frame fusion provided in the embodiments of the present invention. For the sake of brevity, it will not be described in detail here.
[0046] To address the challenge of balancing ranging accuracy, on-chip storage resources, power consumption, and measurement frame rate in existing SPAD ranging technologies, this invention provides a SPAD ranging method and system based on dynamic histogram storage and multi-frame fusion. This method establishes a phase-separated sampling benchmark by introducing multi-frame delay adjustment, performs extreme noise suppression using photon filtering, and utilizes dynamic histogram storage to achieve sparse accumulation only for relevant signal events. Finally, high-precision time-of-flight is extracted through multi-frame data fusion and nonlinear fitting. This method requires only the maintenance of a small amount of valid event information in hardware, significantly reducing storage overhead and power consumption. It also supports flexible switching between measurement accuracy and frame rate through a configurable number of subframes, thereby achieving efficient coordination and optimized balance between ranging accuracy, on-chip resources, power consumption, and frame rate at the system level. This provides an effective technical path for achieving high-resolution 3D imaging under limited hardware resources.
[0047] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.
Claims
1. A SPAD ranging method based on dynamic histogram storage and multi-frame fusion, characterized in that, include: Multi-frame delay adjustment steps: Divide the single-frame measurement period into multiple sub-frames, and configure a corresponding phase delay amount for each sub-frame so that each sub-frame corresponds to a different equivalent sampling time reference; wherein, each sub-frame corresponds to multiple exposure periods; For each subframe, the photon filtering step and the dynamic histogram storage step are performed sequentially, where: The photon filtering step includes: within each exposure cycle, adaptively filtering photon triggering events of multiple SPADs in macropixels by extreme value tracking, and outputting the effective photon response time and corresponding time calibration information and trigger intensity; The dynamic histogram storage step includes: within each exposure cycle, based on the output of the photon screening step in the current exposure cycle, dynamically tracking the return time of the echo photons and generating the corresponding processing result; integrating the processing results of all exposure cycles in the current sub-frame to generate the candidate histogram corresponding to the current sub-frame; Fitting process steps: Based on the candidate histograms corresponding to all subframes and their respective phase delays, a fused histogram with a higher temporal resolution than that of a single subframe is reconstructed through data fusion. Using the time-of-flight extraction algorithm, waveform analysis and nonlinear fitting are performed on the fused histogram to extract the flight time of the echo photons and output the ranging results.
2. The SPAD ranging method based on dynamic histogram storage and multi-frame fusion according to claim 1, characterized in that, The multi-frame delay adjustment step includes: According to the ranging requirements, the single-frame measurement period is divided into multiple sub-frames, and a delay selection signal corresponding to each sub-frame is generated. Multiple delayed signals with different fixed delay values are generated using an open-loop delay chain; Based on the delay selection signal corresponding to a single subframe, one of the multiple delay signals is selected, and its corresponding sampling clock is adjusted so that the sampling time of the subframe is offset from the laser emission time by the corresponding phase delay amount, thereby establishing the equivalent sampling time reference of the subframe.
3. The SPAD ranging method based on dynamic histogram storage and multi-frame fusion according to claim 2, characterized in that, The photon screening step includes: Within the current exposure cycle, the number of photon triggers of multiple SPADs in the macro-pixel is counted within the summation window using extreme value tracking, and this count is used as the trigger intensity at the current sampling moment. Compare the trigger strength at the current sampling time with the current maximum trigger strength; if the trigger strength at the current sampling time is greater, update the current maximum trigger strength using the trigger strength at the current sampling time, and update the current candidate response time using the current sampling time; wherein, the current maximum trigger strength is the maximum number of triggers at the previous sampling time, and the current candidate response time is the sampling time corresponding to the current maximum trigger strength; otherwise, do not update the current maximum trigger strength and the current candidate response time. After the current exposure cycle ends, the current candidate response time is output as the corresponding effective photon response time; The time interval between the effective photon response time and the laser emission time is quantified as the time calibration information.
4. The SPAD ranging method based on dynamic histogram storage and multi-frame fusion according to claim 1 or 3, characterized in that, The dynamic histogram storage step includes: The number of photon triggers corresponding to the effective photon response time output by the photon filtering step within the current exposure cycle is nonlinearly weighted to obtain the weighted trigger intensity. Based on the weighted trigger intensity and the corresponding time calibration information, the distribution correlation between the photon trigger events corresponding to the weighted trigger intensity and the preset event information in the time dimension is analyzed. The trigger time of the photon trigger events that meet the preset correlation conditions is updated as the candidate return time, and the event information in the dynamic histogram storage module is iteratively updated. The preset event information refers to the information of the photon trigger events that meet the correlation conditions stored at the end of the previous exposure cycle. After the current exposure cycle ends, multiple candidate return times are accumulated and stored in order of trigger intensity from high to low to generate the corresponding candidate histogram. The candidate histograms for all exposure periods are aggregated to obtain the candidate histogram for the current subframe.
5. The SPAD ranging method based on dynamic histogram storage and multi-frame fusion according to claim 2, characterized in that, The fitting process includes: Based on the phase delay of each subframe, the data in the corresponding candidate histograms are aligned and fused on the time axis to reconstruct a fused histogram with a time resolution higher than the measurement resolution of any subframe. The waveform distribution characteristics of the fused histogram are analyzed, and the final return time of the echo photon is determined by nonlinear fitting. The final ranging result is obtained by calculating the time difference between the final return time and the laser emission time, combined with the speed of light.
6. The SPAD ranging method based on dynamic histogram storage and multi-frame fusion according to claim 3, characterized in that, The summation window is a dynamic time window, and its window width and / or window position are dynamically determined according to the spatiotemporal distribution characteristics of photon-triggered events in the macropixel.
7. The SPAD ranging method based on dynamic histogram storage and multi-frame fusion according to claim 1, characterized in that, The time calibration information is obtained by quantization through a time-to-digital converter.
8. The SPAD ranging method based on dynamic histogram storage and multi-frame fusion according to claim 2, characterized in that, The open-loop delay chain is composed of multiple cascaded CMOS inverters.
9. The SPAD ranging method based on dynamic histogram storage and multi-frame fusion according to claim 1, characterized in that, The method also includes a working mode switching step: In response to the precision priority command, the number of subframes within the single frame measurement period is increased; In response to the frame rate priority command, the number of subframes within the single frame measurement period is reduced, or the flight time is directly extracted based on the candidate histogram of the single subframe to output a high frame rate ranging result.
10. A SPAD ranging system based on dynamic histogram storage and multi-frame fusion, characterized in that, The system is equipped with the SPAD ranging method according to any one of claims 1 to 9; the system includes: a multi-frame delay adjustment module, a photon filtering module, a dynamic histogram storage module, and a fitting processing module; wherein: The multi-frame delay adjustment module is used to divide a single-frame measurement period into multiple sub-frames and configure a corresponding phase delay amount for each sub-frame so that each sub-frame corresponds to a different equivalent sampling time reference; wherein, each sub-frame corresponds to multiple exposure periods. The photon filtering module and the dynamic histogram storage module process each subframe; wherein: The photon filtering module is used to adaptively filter photon triggering events of multiple SPADs in macropixels in each exposure cycle by means of extreme value tracking, suppress invalid triggering caused by background noise, and output the effective photon response time and corresponding time calibration information and triggering intensity. The dynamic histogram storage module is used to dynamically track the return time of echo photons based on the output of the photon screening step in the current exposure cycle within each exposure cycle, and generate the corresponding processing result; and integrate the processing results of all exposure cycles in the current subframe to generate the candidate histogram corresponding to the current subframe. The fitting processing module is used to reconstruct a fused histogram with a higher temporal resolution than a single subframe by using data fusion based on the candidate histograms corresponding to all subframes and their respective phase delays. The module then uses a time-of-flight extraction algorithm to perform waveform analysis and nonlinear fitting on the fused histogram, extract the flight time of the echo photons, and output the ranging results.