A nonlinear pre-correction method for FMCW lidar based on deep adaptive iteration

Abstract

This paper discloses a method for pre-correcting the nonlinearity of FMCW laser radars based on deep adaptive iteration. This method measures the time-frequency curve of the laser at the current drive voltage, generates a difference signal from the ideal sweep frequency function, and uses an adaptive algorithm to iteratively optimize the drive voltage until the nonlinear error is minimized. This method significantly improves the effective linearized frequency modulation bandwidth and achieves faster convergence. It is suitable for correcting FMCW lasers with varying degrees of nonlinearity and can be used in high-precision, wide-tuning-range FMCW laser radar systems.

Description

Technical Field

[0001] The present invention relates to the technical field of frequency modulated continuous wave laser radar, and in particular to a nonlinear pre-correction method of an FMCW laser radar based on deep adaptive iteration. Background Art

[0002] Frequency-modulated continuous-wave (FWCM) lidar offers advantages such as immunity to ambient light interference, high measurement resolution, and high accuracy. However, its ranging performance depends heavily on the linearity of the laser source's frequency sweep. Due to the inherent characteristics of FM lasers, the relationship between the uncorrected output optical frequency and time is not strictly linear. Therefore, nonlinear pre-correction of FM lasers is necessary in practical applications.

[0003] At present, the nonlinear pre-correction of frequency-modulated lasers is mainly achieved through the current iteration algorithm. That is, by measuring the time-frequency curve of the relationship between the laser frequency and time, combined with the iterative algorithm, the driving voltage signal is updated to make the output light frequency change linearly (1. X. Zhang, J. Pouls, and M. Wu. “Laser frequency sweep linearization by iterative

[0004] learning pre-distortion for FMCW LiDAR." Opt. Express 27(7), 9965-9974 (2019). 2.

[0005] Li P, Zhang YT, Yao J Q. Rapid linear frequency swept frequency-modulated continuous

[0006] wave laser source using iterative pre-distortion algorithm[J]. RemoteSensing, 2022,14(14):

[0007] 3455-3465.). However, since the fixed proportional coefficient cannot be dynamically adjusted according to the error, this method requires multiple iterations or a large number of complex operations during operation, and cannot quickly achieve nonlinear correction. Summary of the Invention

[0008] In view of the shortcomings of the existing technology, the present invention proposes a FMCW lidar nonlinear pre-correction method based on deep adaptive iteration to quickly correct FMCW lasers with different degrees of nonlinearity.

[0009] In order to solve the above technical problems, the technical solution of the present invention is: a nonlinear pre-correction method of FMCW lidar based on deep adaptive iteration, comprising the following steps:

[0010] Step 1: Measure the optical frequency signal output by the FMCW laser under the initial driving voltage signal to obtain the initial time-frequency curve representing the relationship between the optical frequency and time, and calculate the initial residual nonlinearity.

[0011] Step 2: Use the least squares method to fit the current time-frequency curve into a linear function, remove the intercept term of the linear function, and generate an ideal sweep frequency function;

[0012] Step 3: Obtain the nonlinear frequency error signal by calculating the difference between the measured time-frequency curve and the ideal sweep frequency function point by point , record its maximum absolute deviation value ;

[0013] Step 4: Convert the error signal Converted into compensation voltage signal , and adaptively iterate the proportional coefficient according to the following rules :

[0014] Updated scale factor: , is the number of iterations, hour, is the initial proportional coefficient;

[0015] when When the frequency is greater than or equal to 1GHz, [0.1,1];

[0016] when When the frequency is less than 1 GHz, [-0.1,-0.9];

[0017] Step 5: Compensation Voltage Signal superimposed on the current driving voltage signal to generate an updated driving voltage signal;

[0018] Step 6: Measure the time-frequency curve of the laser output under the new driving voltage signal and recalculate the residual nonlinearity;

[0019] Step 7: Repeat steps 1 to 6. When the current residual nonlinear value is greater than the historical minimum value three times in a row, terminate the iteration and output the corresponding historical optimal driving voltage signal.

[0020] Preferably, the time-frequency curve is measured by an unbalanced Mach-Zehnder interferometer.

[0021] Preferably, the driving voltage signal may also be converted into a driving current signal.

[0022] Preferably, the laser driving voltage signal amplitude range is 0-5V.

[0023] Preferably, the driving voltage signal is a triangle wave signal or a sawtooth wave signal.

[0024] As an advantage, the initial proportional coefficient in step 4 is It can be predicted through manual calibration or computer deep learning.

[0025] The present invention has the following characteristics and beneficial effects:

[0026] The present invention dynamically updates the ideal sweep function during the iterative process, optimizes the sweep slope in real time, and improves the effective linearized frequency modulation bandwidth ratio of the laser within the maximum tunable bandwidth. In an embodiment of the present invention, this ratio can reach up to 98%.

[0027] The present invention adopts adaptive proportional coefficient and step length , achieving efficient compensation for nonlinear error signals, making the convergence faster, significantly reducing the number of iterations, and being compatible with FMCW lasers of different degrees of nonlinearity, and able to tolerate initial frequency deviations at the GHz level. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a flowchart of an embodiment of a FMCW lidar nonlinear pre-correction method based on depth adaptive iteration of the present invention.

[0029] Figure 2 3 is a comparison diagram of the driving voltage signal waveforms before and after iteration of an embodiment of the present invention.

[0030] Figure 3 This is a time-frequency curve result diagram before nonlinear pre-correction in an embodiment of the present invention.

[0031] Figure 4 This is a time-frequency curve result diagram after nonlinear pre-correction in an embodiment of the present invention.

[0032] Figure 5 The figure is a comparison chart of the time-frequency curve results before and after nonlinear pre-correction for comparison. DETAILED DESCRIPTION

[0033] The present invention will be further described in detail below with reference to the accompanying drawings and examples, but the scope of protection of the present invention should not be limited thereto.

[0034] The present invention provides a nonlinear pre-correction method for FMCW lidar based on deep adaptive iteration, such as Figure 1 As shown, the following steps are included:

[0035] Step 1: Measure the time-frequency curve of the FMCW laser when driven by a driving voltage signal, and simultaneously calculate the residual nonlinearity at the current moment. The time-frequency curve is used to characterize the relationship between optical frequency and time.

[0036] Specifically, in this embodiment, a triangle wave signal with a modulation frequency of 10 kHz and a voltage amplitude range of 0-1750 mV was input into a signal generator as the initial driving voltage signal. This initial driving voltage signal was then fed into a DFB laser, where the laser temperature was controlled at 25°C. The laser output signal was then fed into an unbalanced Mach-Zehnder interferometer with a 5-meter delay fiber. A phase demodulation algorithm was used to obtain the initial time-frequency curve and calculate the initial residual nonlinearity.

[0037] The calculation method of residual nonlinearity is: first calculate the residual sum of squares of the time-frequency curve, which is used to represent the cumulative deviation of the time-frequency curve from the ideal linear model; calculate the total sum of squares of the time-frequency curve, which is used to represent the total variation of the time-frequency curve relative to its mean; and the ratio of the residual sum of squares to the total sum of squares is the residual nonlinearity (the coefficient of determination in linear regression is obtained by calculating these two items, which is used to measure the linearity of the time-frequency curve. The residual nonlinearity is 1-coefficient of determination).

[0038] Specifically, such as Figure 3 As shown; the residual nonlinearity is , mainly through the linear regression coefficient To determine, that is .in is the residual sum of squares, is the total sum of squares. For a time-frequency curve with N sample points, , .in is the time-frequency curve, is a linear function after time-frequency curve fitting, where the subscript Represents the number of iterative processing in the nonlinear correction process, Time-frequency curve In this embodiment, the residual nonlinearity is .

[0039] Step 2: Use the least squares method to fit the current time-frequency curve into a linear function, remove the intercept term of the linear function, and generate an ideal sweep frequency function. The initial linear function obtained by fitting is , after removing the intercept term, we get the ideal sweep frequency function = .

[0040] Step 3: Obtain the nonlinear frequency error signal by calculating the difference between the measured time-frequency curve and the ideal sweep frequency function point by point , record its maximum absolute deviation value In this embodiment, the maximum absolute deviation value is recorded. It is 2.368GHz.

[0041] It should be noted that It is obtained by point-by-point calculation and cannot be expressed by a specific function.

[0042] Step 4: Transform the error signal Converted into compensation voltage signal , and adaptively iterate the proportional coefficient according to the following rules :

[0043] Updated scale factor: , k is the number of iterations, when k=0, is the initial proportional coefficient, selected based on historical experimental data ;

[0044] when When the frequency is greater than or equal to 1GHz, Take 0.5;

[0045] when When the frequency is less than 1 GHz, Take -0.5.

[0046] Step 5: Compensation Voltage Signal superimposed on the current driving voltage signal to generate an updated driving voltage signal;

[0047] Step 6: Measure the time-frequency curve of the laser output under the new driving voltage signal and recalculate the residual nonlinearity;

[0048] Step 7: Repeat steps 1 to 6. When the current residual nonlinear value is greater than the historical minimum value three times in a row, terminate the iteration and output the corresponding historical optimal driving voltage signal.

[0049] In this embodiment, the initial driving voltage signal and the optimal driving voltage signal after 14 iterations are as follows: Figure 2 As shown in FIG, the iterative change trend of the driving voltage signal can be seen from the figure.

[0050] In this embodiment, after 14 iterations, the up-sweep time-frequency curve and the residual nonlinearity are as follows: Figure 4 As shown. The effective linearized FM bandwidth at this time accounts for 98% of the maximum tunable bandwidth. Figure 3 and Figure 4 , after 14 iterations, the residual nonlinearity changes from Reduce to , where the residual nonlinearity is reduced to 0.0016% of the initial residual nonlinearity after processing.

[0051] Comparative Example

[0052] The iterative learning pre-distortion lidar with frequency-sweep linearization (FMCW) uses the relationship between tuning bandwidth and time (2B / T) to establish a fixed ideal frequency sweep function. It then uses the difference between the time-frequency curve and the ideal sweep function (the difference generates an error signal) for iterative correction. This involves varying the driving voltage signal to alter the time-frequency relationship of the output signal, thereby continuously reducing the nonlinearity of the time-frequency curve. The effective linearized frequency modulation bandwidth accounts for 80% of the maximum tunable bandwidth.

[0053] pass Figure 5 It can be seen that the residual nonlinearity is the After 256 iterations, the residual nonlinearity is reduced from Reduce to .

[0054] By comparing this embodiment with the comparative example, it can be seen that the effective linearized frequency modulation bandwidth of this embodiment is significantly improved as a percentage of the maximum tunable bandwidth.

[0055] Furthermore, after 14 iterations, the residual nonlinearity in this embodiment was reduced to 0.0016% of the initial residual nonlinearity after correction, while the residual nonlinearity in the control example was reduced to 0.0043% of the initial residual nonlinearity after 256 iterations. This demonstrates that the technical solution in this embodiment significantly improves correction performance, requires only a small number of iterations, and significantly improves correction efficiency. This demonstrates that the present invention's real-time nonlinearity correction device and method for FMCW lidar can significantly suppress frequency modulation nonlinearity and improve the ranging accuracy of FMCW lidar.

[0056] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. It will be apparent to those skilled in the art that various changes, modifications, substitutions, and variations of these embodiments, including components, without departing from the principles and spirit of the present invention are still within the scope of protection of the present invention.

Claims

1. A nonlinear pre-correction method for FMCW lidar based on deep adaptive iteration, characterized in that: The steps include: Step 1: measuring the time-frequency curve of the FMCW laser when driven by a driving voltage signal, and simultaneously calculating the residual nonlinearity at the current moment, wherein the time-frequency curve is used to characterize the relationship between optical frequency and time; Step 2: Apply the time-frequency curve to obtain the ideal sweep frequency function, and then calculate the difference between the measured time-frequency curve and the ideal sweep frequency function point by point to obtain the error signal of the nonlinear frequency; Step 3: convert the error signal into a compensation voltage signal, and superimpose the compensation voltage signal on the driving voltage signal in step 1 to generate an updated driving voltage signal; The method of converting the error signal into a compensation voltage signal is: ΔU(t)=P k e(t) Among them, ΔU(t) is the compensation voltage signal at time t, e(t) is the error signal at time t, P k is the adaptive iteration scale coefficient; The iterative method of the adaptive iterative proportional coefficient is: P k+1 =P k (1 + a), where k is the number of iterations. When k = 0, P0 is the initial proportional coefficient. When max|e k (t)| is greater than or equal to 1GHz, a∈[0.1,1]; when max|e k (t)| is less than 1GHz, a∈[-0.1,-0.9], where max|e k (t)| is the maximum absolute deviation value; Step 4: Use the updated driving voltage signal as input to drive the FMCW laser, obtain a new time-frequency curve, and recalculate the residual nonlinearity; Step 5: Repeat steps 1-4. When the current residual nonlinear value is greater than the historical minimum value three times in a row, terminate the iteration and output the corresponding historical optimal driving voltage signal.

2. The method according to claim 1, characterized in that The time-frequency curve is measured by an unbalanced Mach-Zehnder interferometer.

3. The method according to claim 1, characterized in that The residual nonlinearity is calculated by respectively calculating the residual sum of squares and the total sum of squares of the time-frequency curve, and the ratio of the residual sum of squares to the total sum of squares is the residual nonlinearity.

4. The method according to claim 1, wherein The method for obtaining the ideal frequency sweep function is: obtaining a linear function by fitting the time-frequency curve, and removing the intercept term of the linear function to generate the ideal frequency sweep function.

5. The method according to claim 4, characterized in that The fitting method is the least squares method.

6. The method according to claim 1, characterized in that The initial proportional coefficient P0 can be manually calibrated or predicted by a deep learning model based on historical data.

7. The method according to any one of claims 1 to 6, characterized in that The driving voltage signal amplitude range is 0-5V.

8. The method according to any one of claims 1 to 6, characterized in that: The driving voltage signal is a triangle wave signal or a sawtooth wave signal.

9. The method according to any one of claims 1 to 6, characterized in that: The driving voltage signal is replaced by a driving current signal.

Images