A method for crosstalk error compensation of a multi-channel optical transceiver module

Abstract

The application belongs to the technical field of optical communication, and particularly relates to a crosstalk error compensation method of a multi-channel optical transceiver module, which comprises the following steps: collecting the backlight monitoring current of each receiving channel, calculating the total load current estimation of the common bias rail, combining the equivalent output impedance of the common bias rail, and calculating the actual reverse bias voltage of each receiving channel; calculating the current avalanche gain of each receiving channel based on the actual reverse bias voltage, further determining the gain drift rate of each receiving channel relative to the rated gain, and constructing a dynamic crosstalk coefficient matrix based on the gain drift rate and the static crosstalk coefficient; organizing the original output voltage signal of each receiving channel into a column vector, performing linear crosstalk elimination on the column vector by using the dynamic crosstalk coefficient matrix, and obtaining the compensated signal vector. The application eliminates the defect that the fixed crosstalk matrix cannot track the dynamic change of the gain, and improves the decision threshold stability in the burst mode.

Description

Technical Field

[0001] This invention belongs to the field of optical communication technology, and specifically relates to a method for compensating crosstalk error in a multi-channel optical transceiver module. Background Technology

[0002] In the field of high-speed optical communication, multi-channel optical transceiver modules such as 100G CWDM LR4 and 200G / 400G WDM have become core components of data center interconnects and access network aggregation layers. These modules integrate multiple laser transmitting components and photoelectric detection and receiving components operating at different wavelengths into a single package with high density, and the aggregation rate carried by a single module continues to evolve towards the hundreds of gigabits per second. At the same time, composite modules for access networks, such as Comb PON OLT, also integrate multiple independent wavelength transmitting and receiving channels in the same housing, with the physical spacing between channels only on the order of millimeters.

[0003] Patent document CN119853707B discloses a method and apparatus for crosstalk cancellation in multi-channel data transmission based on digital encoding. The method involves the transmitter converting every three parallel signals to four signals for each channel to eliminate crosstalk, and then combining the converted signals before sending them into the channel for data transmission. The receiver, on the other hand, converts every four signals received from each channel back into three signals to obtain the original data. The apparatus includes an encoder for converting the three signals to four signals and a decoder for converting the four signals received by the receiver back into three signals. This invention aims to achieve maximum crosstalk cancellation while ensuring high efficiency in multi-channel digital signal data transmission.

[0004] To suppress optical crosstalk between channels, existing technologies generally employ thin-film interference filters or arrayed waveguide gratings for wavelength division multiplexing isolation. Simultaneously, independent avalanche photodiodes combined with transimpedance amplifiers are deployed in each receiving channel to complete photoelectric conversion. Furthermore, a fixed-coefficient crosstalk cancellation matrix, calibrated offline before shipment, is introduced at the signal processing layer. This matrix subtracts the output signals of the transimpedance amplifiers from adjacent channels by a fixed ratio, thereby eliminating residual optical and electrical crosstalk components. This approach effectively suppresses static crosstalk in continuous wave operation mode.

[0005] However, in multi-channel avalanche photodiode receiver arrays, the reverse high-voltage bias of each avalanche photodiode is typically provided by a common on-chip or module high-voltage bias power rail. When a burst packet input or optical power step occurs in a channel, the transient photocurrent of that channel's avalanche photodiode surges, creating a transient voltage drop across the parasitic impedance of the common bias rail. This causes the actual reverse bias voltages of all other avalanche photodiodes sharing the same bias rail to shift in the same direction at the same moment. Since the avalanche gain of the avalanche photodiode is highly nonlinear in its dependence on the bias voltage, even a small drift in the bias voltage can cause a significant change in the gain value, affecting the crosstalk propagation coefficient between channels and making it a nonlinear quantity that dynamically changes with the instantaneous photocurrent state of each channel. Therefore, the fixed matrix coefficients calibrated at the factory based on static operating conditions cannot track this dynamic change at burst boundary moments, resulting in residual error components positively correlated with the gain drift amplitude after compensation, causing decision threshold drift at the bit error rate-sensitive burst boundary. Summary of the Invention

[0006] This invention provides a crosstalk error compensation method for a multi-channel optical transceiver module to solve the technical problem that existing fixed crosstalk coefficient matrices cannot track the dynamic gain changes of multi-channel avalanche photodiodes caused by transient disturbances in the common bias rail at sudden boundary moments.

[0007] This invention provides a method for compensating crosstalk error in a multi-channel optical transceiver module, comprising the following steps: S1, collect the backlight monitoring current of each receiving channel, calculate the estimated total load current of the common bias rail, and, in combination with the equivalent output impedance of the common bias rail, estimate the actual reverse bias voltage of each receiving channel. S2, calculate the current avalanche gain of each receiving channel based on the actual reverse bias voltage, then determine the gain drift rate of each receiving channel relative to the rated gain, and construct the dynamic crosstalk coefficient matrix based on the gain drift rate and the static crosstalk coefficient. S3, organize the original output voltage signals of each receiving channel into column vectors, and use the dynamic crosstalk coefficient matrix to perform linear crosstalk cancellation on the column vectors to obtain the compensated signal vector; S4 makes a decision on the compensated signal vector and calculates the residual, extracts the residual energy estimate, and triggers the adaptive update process of the characteristic model parameters when the residual energy estimate exceeds the baseline.

[0008] Its effect is as follows: by inverting the avalanche gain state of each channel avalanche photodiode in real time, a crosstalk coefficient matrix that is dynamically updated with bias perturbation is constructed, eliminating the physical root cause that a fixed crosstalk matrix cannot track dynamic changes in gain. While ensuring the elimination of residual compensation error at burst boundaries, the decision threshold drift is reduced, enabling the multi-channel optical transceiver module to maintain stable crosstalk compensation performance in burst mode.

[0009] Furthermore, the total load current estimate of the common bias rail is calculated, and combined with the equivalent output impedance of the common bias rail, the actual reverse bias voltage of each receiving channel is estimated, including: Obtain the real-time monitoring current of each receiving channel, and calculate the sum of the monitoring currents of all receiving channels as the total load current estimate; Multiply the total load current estimate by the equivalent output impedance to obtain the transient voltage drop across the bias rail; The actual reverse bias voltage of each receiving channel is calculated by subtracting the transient voltage drop from the rated operating bias voltage.

[0010] Its effects are as follows: by summarizing the current monitored by all channels to obtain the total load current, and combining it with the equivalent impedance of the bias rail to calculate the transient voltage drop, and then subtracting the voltage drop from the rated bias voltage to obtain the real-time bias voltage, it achieves real-time voltage inversion without additional hardware based on a simple linear Ohm model. It has low computational load and fast response speed. It can synchronously capture nanosecond-level current step disturbances brought by burst data packets, providing an accurate and synchronous voltage input basis for subsequent avalanche gain real-time calculation, and avoiding the attenuation of subsequent crosstalk compensation accuracy caused by bias voltage estimation deviation.

[0011] Furthermore, the steps for obtaining the equivalent output impedance include: Before the bias circuits of each receiving channel are connected, a controlled DC excitation current of known amplitude is output to the bias rail. Read the corresponding change in the bias rail voltage, calculate the ratio of the corresponding change to the controlled DC excitation current, and obtain the equivalent output impedance.

[0012] Furthermore, the current avalanche gain of each receiving channel is calculated based on the actual reverse bias voltage, thereby determining the gain drift rate of each receiving channel relative to its rated gain, including: Obtain the avalanche breakdown voltage of each receiving channel and the empirical power coefficient related to the device material system; Substitute the actual reverse bias voltage, avalanche breakdown voltage, and empirical power coefficient into the gain characteristic formula to calculate the current avalanche gain. Calculate the ratio of the current avalanche gain to the rated gain, and subtract 1 from the ratio to determine the gain drift rate of each receiving channel relative to the rated gain.

[0013] Furthermore, a dynamic crosstalk coefficient matrix is ​​constructed based on the gain drift rate and the static crosstalk coefficient, including: Obtain the first gain drift rate of the interference source channel and the second gain drift rate of the interfered channel; The dynamic interference ratio is calculated by using the value of the first gain drift rate plus 1 as the numerator and the value of the second gain drift rate plus 1 as the denominator. The dynamic interference coefficient is obtained by multiplying the static crosstalk coefficient by the dynamic interference ratio, and a dynamic crosstalk coefficient matrix is ​​constructed based on the dynamic interference coefficients between each channel.

[0014] Its effect is that it provides a more complete physical picture and avoids the problem of insufficient compensation accuracy caused by unilateral correction under extreme conditions where the gain drift directions of the interference source channel and the interference channel are opposite.

[0015] Furthermore, linear crosstalk cancellation is performed on the column vectors using the dynamic crosstalk coefficient matrix to obtain the compensated signal vector, including: Set the diagonal elements of the dynamic crosstalk coefficient matrix to zero, and retain the off-diagonal elements; The elimination matrix is ​​obtained by subtracting the dynamic crosstalk coefficient matrix with its diagonal elements set to zero from the identity matrix. The compensated signal vector is obtained by performing a matrix-vector product operation between the elimination matrix and the column vector.

[0016] Its effect is that it makes the instruction cycle shorter than the full matrix multiplication and addition instruction cycle, and can reduce the processor's computational load while maintaining the same compensation effect, thus adapting to hardware environments with limited computing resources.

[0017] Furthermore, a decision is made on the compensated signal vector and the residual is calculated. The residual energy estimate is then extracted, including: The compensated signal vector is used to make a decision, and the decision symbol for each receiving channel is obtained. The reconstructed signal is obtained by multiplying the decision symbol by the rated decision reference amplitude. The residual is calculated by subtracting the reconstructed signal from the compensated signal vector. The residual mean square values ​​within the observation interval of the statistical window length are extracted and accumulated to obtain the residual energy estimate for each receiving channel.

[0018] Furthermore, the steps for determining the statistical window length include: Obtain the baseline noise standard deviation and the residual mean estimation error. Determine the minimum sample size based on the ratio between the baseline noise standard deviation and the residual mean estimation error, which serves as the lower bound of the statistical window length. Extract the number of symbols corresponding to the maximum burst packet frame length under the current operating rate of the receiving channel, and use it as the upper bound of the statistical window length; Select an integer value between the lower and upper bounds that satisfies the current rate constraint as the statistical window length.

[0019] Furthermore, when the residual energy estimate exceeds the baseline, an adaptive update process for the characteristic model parameters is triggered, including: When the residual energy estimate exceeds the baseline noise root mean square error, calculate the multiple by which the residual energy estimate exceeds the baseline noise root mean square error; The single-step correction amount is calculated based on the multiple, the allowable range of variation of the characteristic model parameters, and the maximum number of convergence iterations. The correction direction is determined by comparing the magnitudes of the residual energy estimates before and after the previous correction step, and the characteristic model parameters are updated by combining the correction direction with the single-step correction amount.

[0020] Its effect is to ensure that the compensation accuracy continues to self-optimize as the module accumulates working time, reduce the limitations of device characteristic snapshots during factory calibration, and enable the multi-channel optical transceiver module to maintain stable crosstalk compensation performance throughout its entire life cycle.

[0021] Furthermore, the backlight monitoring current of each receiving channel is collected, including: Determine the frame boundary time resolution corresponding to the maximum burst rate of the receiving channel; The sampling clock frequency is set based on the frame boundary time resolution, and the backlight monitoring current of each receiving channel is collected according to the sampling clock frequency.

[0022] The beneficial effects are as follows: This invention uses the backlight monitoring current of each channel avalanche photodiode as the native information source. Without adding any external sensors, it reconstructs the gain state of each channel avalanche photodiode at the burst boundary through real-time inversion of the transient voltage drop of the bias rail. Compared with existing fixed coefficient matrices, this scheme reduces the crosstalk compensation residual at the burst boundary, improving the stability of the decision threshold in burst mode.

[0023] The dynamic matrix simultaneously corrects the gain states of both the interference source channel and the interfered channel, achieving a compensation accuracy superior to simplified schemes that only correct one-sided gain. The residual-driven adaptive parameter update path enables the compensation accuracy to continuously track the gradual characteristic changes caused by device aging or temperature drift, extending the module's effective calibration lifespan. The self-calibration process of the equivalent output impedance of the common bias rail during the power-on self-test phase only requires the integration of a built-in digital-to-analog converter and an analog-to-digital converter, resulting in low engineering implementation costs and good batch consistency. Attached Figure Description

[0024] Figure 1 This is a flowchart of the crosstalk error compensation method for the multi-channel optical transceiver module in this invention.

[0025] Figure 2 This is a curve comparison of the crosstalk compensation residual as a function of the signal-to-noise ratio in this invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] An embodiment of the crosstalk error compensation method for a multi-channel optical transceiver module provided by the present invention: like Figure 1 As shown, the crosstalk error compensation method for a multi-channel optical transceiver module includes the following steps: S1, collect the backlight monitoring current to calculate the actual reverse bias voltage.

[0028] In an optional embodiment, the multi-channel optical transceiver module samples the avalanche photodiode backlight monitoring current of each receiving channel in real time during normal operation. Backlight monitoring current of the channel The current monitoring value is determined by the photoelectric conversion responsivity and incident light power of each channel, and is output internally through an independent current monitoring pin or the current mirror circuit built into the transimpedance amplifier. The frame boundary time resolution corresponding to the maximum burst rate of the receiving channel is determined, and the sampling clock frequency is set based on this resolution. The backlight monitoring current of each receiving channel is collected according to the sampling clock frequency to ensure that current step events at the nanosecond to microsecond level can be captured, avoiding missed detections due to under-sampling at burst boundary moments.

[0029] Furthermore, in response to the module's power-on self-test phase, before the avalanche photodiode bias circuits of each channel are connected, a controlled DC excitation current of known amplitude is output to the bias rail via the built-in digital-to-analog converter. Simultaneously, the corresponding change in bias rail voltage is read by the bias monitoring analog-to-digital converter. Calculate the ratio of the corresponding change to the controlled DC excitation current to obtain the equivalent output impedance of the common bias rail. :

[0030] income The parameters are stored in a non-volatile register inside the module and used as fixed parameters throughout the entire operating life. Due to differences in packaging processes, the parasitic impedance of the bias rail varies between different batches of modules. If the nominal design value is used to uniformly replace the measured value, the systematic error will be too large. Therefore, this difference is solidified module by module through a self-calibration process to reduce systematic error.

[0031] Next, after obtaining the real-time monitoring current of each receiving channel, the total current is calculated. The sum of the monitoring currents of each receiving channel is used as an estimate of the total load current of the common bias rail. :

[0032] Multiplying the total load current estimate by the equivalent output impedance, and based on the ohmic voltage drop across the common bias rail impedance, the transient voltage drop across the bias rail is obtained. for:

[0033] Since all avalanche photodiodes are mounted on the same bias rail, the transient voltage drop can be calculated by subtracting the rated operating bias voltage. The passage is Actual reverse bias voltage at time for:

[0034] in This is the rated operating bias voltage set at the time of module factory calibration.

[0035] Using the sampling period as the time interval, the actual instantaneous reverse bias voltage of each channel avalanche photodiode is restored through pure arithmetic operations by summing the one-time impedance self-calibration during the power-on self-test stage and the backlight monitoring current during operation, and providing real-time input for subsequent gain state calculation. The inversion delay is strictly synchronized with the sampling period.

[0036] S2, calculate the avalanche gain to construct the dynamic crosstalk coefficient matrix.

[0037] In an optional embodiment, the avalanche breakdown voltage of each receiving channel and the empirical power coefficient related to the device material system are obtained; the actual reverse bias voltage of each channel, which has already been inverted, is then used. Substituting the avalanche breakdown voltage and empirical power coefficients into the Emmons formula, which describes the gain characteristics of an avalanche photodiode, the gain characteristic formula is calculated in real time to obtain the first... Current avalanche gain at the current moment of the channel :

[0038] in For the first Avalanche breakdown voltage of a channel avalanche photodiode These are empirical power coefficients related to the device material system. For InGaAs / InP avalanche photodiodes, The typical value range is 3 to 5. Both are determined by gain-voltage curve fitting and stored in the module's internal non-volatile register before the module leaves the factory.

[0039] Then, the ratio of the current avalanche gain to the rated gain is calculated, and 1 is subtracted from the ratio to determine the rated gain of each receiving channel relative to the factory calibration time. Gain drift rate Defined as:

[0040] in, This value represents the relative offset of the transimpedance amplifier output signal amplitude relative to the calibration state. A larger value indicates a potentially larger deviation between the actual current bias voltage and the rated bias voltage, resulting in more significant dynamic drift in the channel gain; a smaller value indicates potentially smaller bias rail disturbances, resulting in a gain closer to the factory calibration state. When there is no bias rail disturbance... The subsequent matrix automatically degenerates into the factory static calibration matrix, which is fully compatible with existing technologies.

[0041] Next, based on the gain drift rate and static crosstalk coefficient of each channel, a system is constructed. dimensional dynamic crosstalk coefficient matrix Obtain the first gain drift rate of the interference source channel and the second gain drift rate of the interfered channel; calculate the dynamic interference ratio by adding one to the first gain drift rate as the numerator and adding one to the second gain drift rate as the denominator; multiply the static crosstalk coefficient by the dynamic interference ratio to obtain the dynamic interference coefficient, and construct a dynamic crosstalk coefficient matrix based on the dynamic interference coefficients between each channel. Matrix elements This represents the dynamic interference coefficient caused by the crosstalk signal of channel j to channel i. Its numerator is proportional to the actual output amplitude of channel j, and its denominator is proportional to the current full-scale reference amplitude of channel i. Therefore, the dynamic coefficient is:

[0042] in The static crosstalk coefficient, calibrated by isolation test under rated bias and rated gain conditions at the factory, is stored in the non-volatile register inside the module. and The first number is calculated in real time. The first gain drift rate of the channel, i.e., the interference source channel, and the second... The channel is the second gain drift rate of the interfered channel. It should be noted that there are many models in this field that describe the gain characteristics of avalanche photodiodes. Those skilled in the art can choose different gain-voltage relationships according to the actual device material system and calibration accuracy requirements. As long as the bias voltage obtained by inversion can be mapped to the gain drift rate, the matrix can be constructed according to the above dynamic coefficient relationship.

[0043] Compared to the simplified scheme that only corrects the gain on one side, this scheme provides a more complete physical picture under extreme conditions where the gain drift directions of the interference source channel and the interfered channel are opposite, thus avoiding the problem of insufficient compensation accuracy caused by one-sided correction.

[0044] In this way, the dynamic crosstalk coefficient matrix is ​​updated periodically on a sampling basis based on the real-time gain drift rate of each channel, so that the matrix coefficients are always synchronized with the current offset track disturbance state, providing accurate dynamic coefficient input for subsequent linear crosstalk cancellation.

[0045] S3 utilizes a dynamic matrix to perform linear crosstalk cancellation.

[0046] In an optional embodiment, the raw output voltage signals of the transimpedance amplifiers of each receiving channel are organized into column vectors. ,in Includes valid signals from this channel and signals from the rest. The crosstalk superposition components of each channel. For each sampling time, using the pre-constructed dynamic crosstalk coefficient matrix... Perform linear crosstalk cancellation on the column vector to obtain the compensated signal vector:

[0047] in for An identity matrix of order 1. This is the estimated effective signal vector after crosstalk compensation for each channel. The diagonal elements of the dynamic crosstalk coefficient matrix are set to zero, while the off-diagonal elements are retained. Since each channel does not create crosstalk with itself, only the off-diagonal elements participate in the calculation. The elimination matrix is ​​obtained by subtracting the dynamically crosstalk coefficient matrix with zeroed diagonal elements from the identity matrix. The compensation signal vector is obtained by performing a matrix-vector product operation between the elimination matrix and the column vectors. By replacing full matrix operations with setting the diagonal elements to zero, the instruction cycle is significantly shorter than the full matrix multiplication and addition instruction cycle, reducing the processor's computational load while maintaining the same compensation effect.

[0048] Furthermore, the compensation operation is executed in a pipelined manner within the digital signal processing unit inside the module, and the matrix-vector product operation is implemented through a parallel multiply-accumulate logic array. In burst mode applications, taking the concurrent operation of the 1270nm and 1310nm dual avalanche photodiode receiving channels in a Comb PON OLT as an example, in response to a high-power burst packet received by one channel, the backlight monitoring current of that channel... The transient step is captured within the next sampling period, corresponding to the channel's gain drift rate. The dynamic crosstalk coefficient matrix is ​​updated accordingly. Adaptive switching is achieved. Compared to the fixed coefficients of the static matrix at the burst boundary, this approach provides a gradual transition when the gain state changes, avoiding the problem of the residual surging near the burst boundary due to the gain state change.

[0049] Compensated signals of each channel It is then fed into the subsequent decision circuit.

[0050] Using the backlight monitoring current sampling period as the time interval, dynamic matrix-vector product operations are performed through a pipelined parallel multiply-accumulate logic array to control the crosstalk residual at the sudden boundary within the single sampling period convergence range after the bias disturbance disappears.

[0051] S4, calculate the residuals to trigger the parameter adaptive update process.

[0052] In an optional embodiment, a decision is made on the compensated signal vector to obtain the decision symbol for each receiving channel. The reconstructed signal is obtained by multiplying the decision symbol by the nominal decision reference amplitude. The residual between the pre-decision signal and the post-decision reconstructed signal is calculated by subtracting the reconstructed signal from the compensated signal vector.

[0053] in For the first The rated decision reference range of the channel. This reflects the amount of residual error after compensation. The statistical window length is [length missing]. The residual mean square values ​​within the observation intervals of each symbol are accumulated to obtain the residual energy estimate for each receiving channel. :

[0054] Next, the baseline noise standard deviation and the residual mean estimation error are obtained. Based on the ratio between the baseline noise standard deviation and the residual mean estimation error, the minimum sample size is determined as the lower bound of the statistical window length. The number of symbols corresponding to the maximum burst packet frame length at the current operating rate of the receiving channel is extracted as the upper bound of the statistical window length. An integer value that satisfies the current rate constraint is selected between the lower and upper bounds as the statistical window length. Based on the required accuracy of the residual mean estimation, the estimation error is determined. relative to baseline noise standard deviation The preset ratio is used to obtain the minimum sample size. For example, when the preset ratio is one-tenth, The value can be 100. This is the baseline noise root mean square error measured and stored during the power-on self-test phase of the module's receiving channel under no-load conditions. This method of value determination synchronously binds the statistical window with the module's true noise floor and frame structure.

[0055] Residual energy estimator for any channel Exceeding the upper bound of acceptable residuals Trigger the MV feature model parameters of this channel The adaptive update process is as follows: When the residual energy estimate exceeds the baseline noise mean square error, the multiple by which the residual energy estimate exceeds the baseline noise mean square error is calculated; based on the multiple, the allowable range of variation of the characteristic model parameters, and the maximum number of convergence iterations, the single-step correction amount is calculated. It is determined by the following arithmetic expression:

[0056] in for The allowable variation range width, due to the InGaAs / InP material system The typical range is 3 to 5, so the interval width is 2; The maximum number of convergence iterations, determined by observing the residual convergence process during the module's power-on self-test phase, is stored in the module's non-volatile register. The single-step correction amount is positively correlated with the factor by which the residual exceeds the baseline; the larger the residual exceeds the baseline, the more severe the deviation in device characteristics, and the larger the factor, the greater the amplification of the step size, resulting in faster convergence. The upper bound of the step size is directly determined by the device's physical constraints. The correction direction is determined by comparing the magnitudes of the residual energy estimates before and after the previous correction step, and the characteristic model parameters are updated by combining the correction direction with the single-step correction amount. The correction direction is determined by comparing the magnitudes of the residual energy estimates before and after the previous correction step. Size comparison determination: when increasing A positive sign is used when the gain estimate is closer to the true gain driven by the residual, and a negative sign is used otherwise.

[0057] The updated Write back to the module's internal non-volatile registers, and the changes will take effect immediately in subsequent gain drift rate calculations. Since the gradual changes in MV characteristics caused by aging and temperature drift occur on a timescale much longer than the update cycle of the residual statistics window, the long-term evolution trend of device characteristics, if not tracked, will continuously accumulate compensation errors, shortening the module's effective calibration lifespan.

[0058] In this way, by driving the continuous adaptive update of MV characteristic parameters through residual energy monitoring, the compensation accuracy is continuously optimized with the cumulative working time of the module, effectively reducing the limitations of the factory-calibrated time device characteristic snapshot, and enabling the multi-channel optical transceiver module to maintain stable crosstalk compensation performance throughout its entire life cycle.

[0059] like Figure 2As shown, the area between the two curves is filled with sparse dotted patterns, visually representing the residual difference distribution of the two methods under different signal-to-noise ratio conditions. It can be observed that both curves show a downward trend as the signal-to-noise ratio increases. The curve corresponding to the method of the present invention is generally below the curve of the existing method, indicating that the method of the present invention has lower crosstalk compensation residuals at all signal-to-noise ratio levels, and the compensation effect is more stable.

[0060] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for compensating crosstalk error in a multi-channel optical transceiver module, characterized in that, Includes the following steps: S1, collect the backlight monitoring current of each receiving channel, calculate the estimated total load current of the common bias rail, and, in combination with the equivalent output impedance of the common bias rail, estimate the actual reverse bias voltage of each receiving channel. S2, calculate the current avalanche gain of each receiving channel based on the actual reverse bias voltage, then determine the gain drift rate of each receiving channel relative to the rated gain, and construct the dynamic crosstalk coefficient matrix based on the gain drift rate and the static crosstalk coefficient. S3, organize the original output voltage signals of each receiving channel into column vectors, and use the dynamic crosstalk coefficient matrix to perform linear crosstalk cancellation on the column vectors to obtain the compensated signal vector; S4 makes a decision on the compensated signal vector and calculates the residual, extracts the residual energy estimate, and triggers the adaptive update process of the characteristic model parameters when the residual energy estimate exceeds the baseline.

2. The crosstalk error compensation method for a multi-channel optical transceiver module according to claim 1, characterized in that, Calculate the estimated total load current of the common bias rail, and based on the equivalent output impedance of the common bias rail, estimate the actual reverse bias voltage of each receiving channel, including: Obtain the real-time monitoring current of each receiving channel, and calculate the sum of the monitoring currents of all receiving channels as the total load current estimate; Multiply the total load current estimate by the equivalent output impedance to obtain the transient voltage drop across the bias rail; The actual reverse bias voltage of each receiving channel is calculated by subtracting the transient voltage drop from the rated operating bias voltage.

3. The crosstalk error compensation method for a multi-channel optical transceiver module according to claim 2, characterized in that, The steps for obtaining the equivalent output impedance include: Before the bias circuits of each receiving channel are connected, a controlled DC excitation current of known amplitude is output to the bias rail. Read the corresponding change in the bias rail voltage, calculate the ratio of the corresponding change to the controlled DC excitation current, and obtain the equivalent output impedance.

4. The crosstalk error compensation method for a multi-channel optical transceiver module according to claim 1, characterized in that, The current avalanche gain of each receiving channel is calculated based on the actual reverse bias voltage, and then the gain drift rate of each receiving channel relative to the rated gain is determined, including: Obtain the avalanche breakdown voltage of each receiving channel and the empirical power coefficient related to the device material system; Substitute the actual reverse bias voltage, avalanche breakdown voltage, and empirical power coefficient into the gain characteristic formula to calculate the current avalanche gain. Calculate the ratio of the current avalanche gain to the rated gain, and subtract 1 from the ratio to determine the gain drift rate of each receiving channel relative to the rated gain.

5. The crosstalk error compensation method for a multi-channel optical transceiver module according to claim 1, characterized in that, A dynamic crosstalk coefficient matrix is ​​constructed based on the gain drift rate and the static crosstalk coefficient, including: Obtain the first gain drift rate of the interference source channel and the second gain drift rate of the interfered channel; The dynamic interference ratio is calculated by using the value of the first gain drift rate plus 1 as the numerator and the value of the second gain drift rate plus 1 as the denominator. The dynamic interference coefficient is obtained by multiplying the static crosstalk coefficient by the dynamic interference ratio, and a dynamic crosstalk coefficient matrix is ​​constructed based on the dynamic interference coefficients between each channel.

6. The crosstalk error compensation method for a multi-channel optical transceiver module according to claim 1, characterized in that, Linear crosstalk cancellation is performed on the column vector using the dynamic crosstalk coefficient matrix to obtain the compensated signal vector, including: Set the diagonal elements of the dynamic crosstalk coefficient matrix to zero, and retain the off-diagonal elements; The elimination matrix is ​​obtained by subtracting the dynamic crosstalk coefficient matrix with its diagonal elements set to zero from the identity matrix. The compensated signal vector is obtained by performing a matrix-vector product operation between the elimination matrix and the column vector.

7. The crosstalk error compensation method for a multi-channel optical transceiver module according to claim 1, characterized in that, The compensated signal vector is used to make a decision and calculate the residual, and the residual energy estimate is extracted, including: The compensated signal vector is used to make a decision, and the decision symbol for each receiving channel is obtained. The reconstructed signal is obtained by multiplying the decision symbol by the rated decision reference amplitude. The residual is calculated by subtracting the reconstructed signal from the compensated signal vector. The residual mean square values ​​within the observation interval of the statistical window length are extracted and accumulated to obtain the residual energy estimate for each receiving channel.

8. The crosstalk error compensation method for a multi-channel optical transceiver module according to claim 7, characterized in that, The steps for determining the statistical window length include: Obtain the baseline noise standard deviation and the residual mean estimation error. Determine the minimum sample size based on the ratio between the baseline noise standard deviation and the residual mean estimation error, which serves as the lower bound of the statistical window length. Extract the number of symbols corresponding to the maximum burst packet frame length under the current operating rate of the receiving channel, and use it as the upper bound of the statistical window length; Select an integer value between the lower and upper bounds that satisfies the current rate constraint as the statistical window length.

9. The crosstalk error compensation method for a multi-channel optical transceiver module according to claim 1, characterized in that, An adaptive update process for the characteristic model parameters is triggered when the residual energy estimate exceeds the baseline, including: When the residual energy estimate exceeds the baseline noise root mean square error, calculate the multiple by which the residual energy estimate exceeds the baseline noise root mean square error; The single-step correction amount is calculated based on the multiple, the allowable range of variation of the characteristic model parameters, and the maximum number of convergence iterations. The correction direction is determined by comparing the magnitudes of the residual energy estimates before and after the previous correction step, and the characteristic model parameters are updated by combining the correction direction with the single-step correction amount.

10. The crosstalk error compensation method for a multi-channel optical transceiver module according to claim 1, characterized in that, Collect the backlight monitoring current of each receiving channel, including: Determine the frame boundary time resolution corresponding to the maximum burst rate of the receiving channel; The sampling clock frequency is set based on the frame boundary time resolution, and the backlight monitoring current of each receiving channel is collected according to the sampling clock frequency.

Images