Dispersion and non-linear compensation method for data center optical interconnection

Abstract

The invention discloses a dispersion and nonlinear compensation method for optical interconnection of a data center, which belongs to the technical field of optical fiber communication, and comprises the following steps: pruning a tap coefficient of a decision feedback equalizer in a second feedforward-decision feedback equalizer based on an absolute value kernel; using the pruned tap coefficient to construct an absolute value kernel-based Thomlinson-Harzimat precoding, and using the absolute value kernel-based Thomlinson-Harzimat precoding to compensate a pulse amplitude modulation signal to be transmitted; according to the method, an absolute value is used as a nonlinear kernel function, the number of multipliers is reduced by half compared with a traditional nonlinear kernel function, meanwhile, the threshold pruning step is introduced, the number of the needed multipliers is further reduced, and therefore dispersion and nonlinear damage in the medium and long distance transmission process can be compensated with low calculation complexity; and the method is suitable for application scenes such as data center optical interconnection with sensitive cost and power consumption and the like.

Description

Technical Field

[0001] The present invention belongs to the technical field of optical fiber communication, and more specifically, relates to a dispersion and nonlinear compensation method for optical interconnection in a data center. Background Art

[0002] With the surge in bandwidth-intensive services such as video streaming, cloud computing, the Internet of Things, and virtual and augmented reality, the continued growth of data center traffic has driven the demand for medium- and long-distance Ethernet exceeding 100Gb / s / λ for data center interconnection. The sensitivity of data center scenarios to cost and power consumption makes the intensity modulation-direct detection (IM / DD) system based on the 4-level pulse amplitude modulation (PAM-4) format the first choice. Double-sideband signals are easily generated by intensity modulation using Mach-Zehnder modulators or directly modulating lasers. However, the power fading effect caused by dispersion in C-band transmission can cause severe distortion of double-sideband signals. Using optical compensation methods such as optical bandpass filters or dispersion-compensating fibers will increase the attenuation of the link and shorten the transmission distance. Therefore, applying dispersion and nonlinear compensation algorithms in digital signal processing at transmitters and receivers is a better choice.

[0003] In the prior art, the dispersion pre-compensation algorithm based on Tomlinson-Halassima precoding (THP) can effectively compensate for the dispersion damage and extend the transmission distance. Due to the nonlinear damage caused by the dispersion channel response and modulator amplifier and other devices, adding a nonlinear kernel to the feedback loop transfer function of THP to become a nonlinear THP can improve the dispersion and nonlinear compensation capabilities and thus improve the system performance. The traditional nonlinear THP uses a nonlinear kernel based on the Volterra series. Compared with the linear THP, the number of multipliers required for the algorithm implementation is greatly increased, which greatly increases the hardware resource cost and energy consumption required for digital signal processing.

[0004] In summary, there is an urgent need for a dispersion and nonlinear compensation method for IM / DD systems with low computational complexity for medium and long-distance data center optical interconnection. Summary of the invention

[0005] In view of the above defects or improvement needs of the prior art, the present invention provides a dispersion and nonlinear compensation method for data center optical interconnection, which aims to solve the technical problem of high calculation complexity of dispersion and nonlinear compensation for medium and long distance data center optical interconnection in the prior art.

[0006] To achieve the above object, according to one aspect of the present invention, a dispersion and nonlinear compensation method for data center optical interconnection is provided, comprising:

[0007] S1: Taking the initial pulse amplitude modulation signal sequence sent by the signal transmitting end as the reference signal and the target pulse amplitude modulation signal sequence recovered by the signal receiving end as the input signal, the first feedforward-decision feedback equalizer based on the absolute value kernel is trained by adjusting the tap coefficient of the equalizer through an adaptive algorithm to obtain the second feedforward-decision feedback equalizer based on the absolute value kernel with the best equalization performance under the transmission link between the signal transmitting end and the signal receiving end;

[0008] S2: pruning tap coefficients of a decision feedback equalizer in the second feedforward-decision feedback equalizer based on the absolute value kernel, and constructing a Tomlinson-Harasima precoding based on the absolute value kernel by using the pruned tap coefficients;

[0009] S3: Compensate the pulse amplitude modulation signal to be transmitted by using the absolute value kernel-based Tomlinson-Harasima precoding to eliminate the dispersion and nonlinear damage caused by the transmission link to the transmission signal.

[0010] In one embodiment, the S2 includes:

[0011] The nonlinear feedback tap coefficients of the decision feedback equalizer in the second feedforward-decision feedback equalizer based on the absolute value kernel are pruned, and the nonlinear feedback tap coefficients greater than a threshold are selected to construct Tomlinson-Harasima precoding based on the absolute value kernel.

[0012] In one of the embodiments, the second feedforward-decision feedback equalizer based on the absolute value kernel includes a second feedforward equalizer and a second decision feedback equalizer, and the second feedforward-decision feedback equalizer based on the absolute value kernel is expressed as:

[0013]

[0014] Wherein, n is the ordinal number of the output signal, k and q both represent the ordinal numbers of the input signals, y(n) represents the nth output signal of the second feedforward-decision feedback equalizer based on the absolute value kernel, w1(k) represents the linear feedforward tap coefficient of the second feedforward equalizer when the kth input signal is input, w2(k,q) represents the nonlinear feedforward tap coefficient of the second feedforward equalizer when the kth input signal and the qth input signal are input, r(·) represents the equalizer input signal sequence, h1(k) represents the linear feedback tap coefficient of the second decision feedback equalizer, h2(k,q) represents the nonlinear feedback tap coefficient of the second decision feedback equalizer, d(·) represents the decision symbol sequence corresponding to the output signal sequence y(n); N1 represents the linear feedforward memory length, N2 represents the nonlinear feedforward memory length, D1 represents the linear feedback memory length, and D2 represents the nonlinear feedback memory length. Indicates a floor operation.

[0015] In one embodiment, the absolute value kernel-based Tomlinson-Halassima precoding constructed in S2 is expressed as:

[0016]

[0017] x(n)=mod 2M (x'(n));

[0018] Among them, x(n), x(nk) and x(nq) represent the nth, nkth and nqth output signals of the second step output sequence of the encoder, x′(n) is the nth output signal of the first step output sequence of the encoder, s(n) represents the nth input symbol of the first step of the encoder; h′2 represents the nonlinear feedback tap coefficient after pruning, mod represents modular operation, and M is the number of levels of the pulse amplitude modulation signal.

[0019] In one embodiment, the S2 includes:

[0020] The linear feedback tap coefficients and the nonlinear feedback tap coefficients of the second feedforward-decision feedback equalizer based on the absolute value kernel are pruned, and the linear feedback tap coefficients whose absolute values ​​are greater than a first threshold and the nonlinear feedback tap coefficients whose absolute values ​​are greater than a second threshold are selected to construct Tomlinson-Harasima precoding based on the absolute value kernel.

[0021] In one embodiment, before S1, the method further includes:

[0022] Generate an original pulse amplitude modulation signal sequence by using the signal transmitting end, and perform up-sampling and pulse shaping on the original pulse amplitude modulation signal sequence to obtain the initial pulse amplitude modulation signal sequence; modulate the initial pulse amplitude modulation signal sequence onto an optical carrier through intensity modulation and transmit it via the transmission link;

[0023] The signal receiving end is used to convert the received optical signal into an electrical signal, and then down-sampling, clock recovery and matched filtering are performed to obtain the target pulse amplitude modulation signal.

[0024] According to another aspect of the present invention, there is provided a communication method for optical interconnection of a data center, comprising:

[0025] A1: At the signal transmitting end, the pulse amplitude modulation signal to be transmitted is compensated by using the dispersion and nonlinear compensation method for data center optical interconnection, and modulated to an optical carrier for transmission;

[0026] A2: At the signal receiving end, the received optical signal is converted into an initial electrical signal, and then down-sampled, clock recovered and matched filtered to obtain a target electrical signal, and then the target electrical signal is decoded to restore the original pulse amplitude modulation signal.

[0027] In one embodiment, the A2 includes: at the signal receiving end: converting the received optical signal into an initial electrical signal, and then performing downsampling, clock recovery and matched filtering to obtain a target electrical signal; using a third feedforward equalizer based on an absolute value kernel and a modular operation to perform Tomlinson-Harasima precoding decoding based on an absolute value kernel on the target electrical signal to restore the original pulse amplitude modulation signal; and demapping the pulse amplitude modulation signal;

[0028] Among them, the third feedforward equalizer based on absolute value kernel is obtained by taking the original pulse amplitude modulation signal s(n) and the first and second step output sequences x′(n) and x(n) after Tomlinson-Harasima precoding based on absolute value kernel, and the result of s(n)+x′(n)-x(n) operation as reference signals, and taking the target pulse amplitude modulation signal sequence recovered by the signal receiving end as input signal, and adjusting the tap coefficients of the second feedforward equalizer based on absolute value kernel through adaptive algorithm for training.

[0029] In one of the embodiments, the third feedforward equalizer based on absolute value kernel is expressed as:

[0030]

[0031] Wherein, k and q both represent the ordinal number of the input signal, y(n) represents the nth output signal of the third feedforward-decision feedback equalizer based on the absolute value kernel, w1(k) represents the linear feedforward tap coefficient of the third feedforward equalizer when the kth input signal is input, w2(k,q) represents the nonlinear feedforward tap coefficient of the third feedforward equalizer when the kth input signal and the qth input signal are input, r(·) represents the equalizer input signal sequence, N1 represents the linear feedforward memory length, N2 represents the nonlinear feedforward memory length, Indicates a floor operation.

[0032] According to another aspect of the present invention, a data center optical interconnection system is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0033] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0034] (1) The present invention provides a dispersion and nonlinearity compensation method for data center optical interconnection, which prunes the tap coefficients of the decision feedback equalizer in the second feedforward-decision feedback equalizer based on the absolute value kernel, and uses the pruned tap coefficients to construct a Tomlinson-Halassima precoding based on the absolute value kernel; the present invention uses the absolute value as the nonlinear kernel function, which reduces the number of multipliers by half compared with the traditional nonlinear kernel based on the Volterra series, and introduces the threshold pruning step to further reduce the number of multipliers required, so that the dispersion and nonlinear damage suffered during the medium and long distance transmission process can be compensated with lower computational complexity, thereby improving the quality of the received signal, and is suitable for application scenarios such as data center optical interconnection that is sensitive to cost and power consumption.

[0035] (2) This scheme prunes the nonlinear feedback tap coefficients of the decision feedback equalizer in the second feedforward-decision feedback equalizer based on the absolute value kernel, selects the nonlinear feedback tap coefficients greater than the threshold and the complete linear feedback tap coefficients to construct the Tomlinson-Halassima precoding based on the absolute value kernel, so that the computational complexity of the subsequent compensation algorithm is lower.

[0036] (3) The second feedforward-decision feedback equalizer based on the absolute value kernel described in this solution is expressed as:

[0037]

[0038] It takes into account the linear and nonlinear signal impairments caused by dispersion and nonlinear effects in the IM / DD transmission system, realizes adaptive modeling of nonlinear channels, and provides the coefficients of the absolute value kernel-based Tomlinson-Harasima precoding.

[0039] (4) The Tomlinson-Halassima precoding based on the absolute value kernel in this scheme is:

[0040]

[0041] It takes into account the linear and nonlinear signal impairments caused by dispersion and nonlinear effects in the IM / DD transmission system, and realizes pre-compensation of the feedback part of the linear and nonlinear signal impairments.

[0042] (5) This scheme selects nonlinear feedback tap coefficients whose absolute values ​​are greater than a first threshold and linear feedback tap coefficients whose absolute values ​​are greater than a second threshold to construct Tomlinson-Halassima precoding based on an absolute value kernel, so that the computational complexity of the subsequent compensation algorithm is lower.

[0043] (6) This solution generates an original pulse amplitude modulation signal sequence at the signal transmitting end and performs up-sampling and pulse shaping on the original pulse amplitude modulation signal sequence to obtain the initial pulse amplitude modulation signal sequence, thereby removing redundant signals.

[0044] (7) The present invention also provides a communication method for data center optical interconnection, which uses pruned tap coefficients to construct a Tomlinson-Halassima precoding based on an absolute value kernel; the Tomlinson-Halassima precoding based on an absolute value kernel is used to compensate for the pulse amplitude modulation signal to be transmitted, so as to eliminate the dispersion and nonlinear damage caused by the transmission link to the transmission signal, and modulate it to an optical carrier for transmission. The dispersion and nonlinear damage suffered during medium and long distance transmission can be compensated with lower computational complexity, thereby improving communication efficiency.

[0045] (8) This scheme uses a third feedforward equalizer based on an absolute value kernel and modular operations to perform absolute value kernel-based Tomlinson-Halassima precoding and decoding on the target electrical signal, which takes into account the linear and nonlinear feedforward signal impairments in the IM / DD transmission system and realizes channel equalization and decoding of the absolute value kernel-based Tomlinson-Halassima precoded signal.

[0046] (9) The third feedforward-decision feedback equalizer based on the absolute value kernel described in this solution is expressed as:

[0047] The linear and nonlinear feedforward signal impairments in the IM / DD transmission system are considered, and post-compensation of the feedforward part of the linear and nonlinear signal impairments is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 is a flow chart of a dispersion and nonlinear compensation method for data center optical interconnection provided by Embodiment 1 of the present invention;

[0049] Figure 2 is a schematic diagram of the structure of Tomlinson-Halassima precoding based on absolute value kernel provided in Example 1 of the present invention;

[0050] Figure 3 is a schematic diagram of the structure of the intensity modulation / direct detection transmission system provided in Example 1 of the present invention;

[0051] Figure 4 is a flow chart of a communication method for data center optical interconnection provided by Embodiment 2 of the present invention;

[0052] Figure 5 is a flow chart of another communication method for data center optical interconnection provided by Embodiment 2 of the present invention;

[0053] Figure 6 This is a comparison chart of bit error rates of the communication method for data center optical interconnection provided by the present invention under different received optical powers, provided by Example 2 of the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0055] Example 1

[0056] like Figure 1 As shown, this embodiment provides a dispersion and nonlinear compensation method for data center optical interconnection, including: S1-S3. S1: Taking the initial pulse amplitude modulation signal sequence sent by the signal transmitting end as the reference signal, and taking the target pulse amplitude modulation signal sequence recovered by the signal receiving end as the input signal, the equalizer tap coefficients are adjusted by an adaptive algorithm to train the first feedforward-decision feedback equalizer based on the absolute value core, and the second feedforward-decision feedback equalizer based on the absolute value core with the best equalization performance under the transmission link between the signal transmitting end and the signal receiving end is obtained; S2: Pruning the tap coefficients of the decision feedback equalizer in the second feedforward-decision feedback equalizer based on the absolute value core, and using the pruned tap coefficients to construct Tomlinson-Halassima precoding based on the absolute value core, such as Figure 2 As shown; S3: Tomlinson-Halassima precoding based on absolute value kernel is used to compensate the pulse amplitude modulation signal to be transmitted to eliminate the dispersion and nonlinear damage caused by the transmission link to the transmission signal.

[0057] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0058] As a preferred implementation, S2 includes: pruning the nonlinear feedback tap coefficients of the decision feedback equalizer in the second feedforward-decision feedback equalizer based on the absolute value kernel, and selecting the nonlinear feedback tap coefficients greater than the threshold to construct Tomlinson-Harasima precoding based on the absolute value kernel.

[0059] As a preferred implementation, the second feedforward-decision feedback equalizer based on the absolute value kernel includes a second feedforward equalizer and a second decision feedback equalizer. The second feedforward-decision feedback equalizer based on the absolute value kernel is expressed as:

[0060]

[0061] Wherein, n is the ordinal number of the output signal, k and q both represent the ordinal numbers of the input signals, y(n) represents the nth output signal of the second feedforward-decision feedback equalizer based on the absolute value kernel, w1(k) represents the linear feedforward tap coefficient of the second feedforward equalizer when the kth input signal is input, w2(k, q) represents the nonlinear feedforward tap coefficient of the second feedforward equalizer when the kth input signal and the qth input signal are input, r(·) represents the equalizer input signal sequence, h(k) represents the linear feedback tap coefficient of the second decision feedback equalizer, h2(k, q) represents the nonlinear feedback tap coefficient of the second decision feedback equalizer, d(·) represents the decision symbol sequence corresponding to the output signal sequence y(n); N1 represents the linear feedforward memory length, N2 represents the nonlinear feedforward memory length, D1 represents the linear feedback memory length, and D2 represents the nonlinear feedback memory length. Indicates a floor operation.

[0062] For example, r is the received signal sampled at twice the baud rate, d is the symbol after decision, and y is the equalizer output; h and w are the decision feedback tap and feedforward tap coefficients; N1=51, N2=35, D1=25, D2=15 are the linear feedforward memory length, nonlinear feedforward memory length, linear feedback memory length, and nonlinear feedback memory length of the channel respectively; Indicates a rounding down operation. The least mean square (LMS) algorithm can be used to train the tap coefficients.

[0063] For example, the obtained nonlinear decision feedback tap h2 is subjected to threshold pruning operation, and the nonlinear term h2(k,q)|x(nk)+x(nq)| whose absolute value of the tap coefficient |h2(k,q)| is less than the threshold value T=0.006 is discarded. The pruned decision feedback tap is used to construct Tomlinson-Halassima precoding based on the absolute value kernel.

[0064] As a preferred implementation, the absolute value kernel-based Tomlinson-Halassima precoding constructed in S2 is expressed as:

[0065]

[0066] x(n)=mod 2M (x′(n));

[0067] Among them, x(n), x(nk) and x(nq) represent the nth, nkth and nqth output signals of the second step output sequence of the encoder, x′(n) is the nth output signal of the first step output sequence of the encoder, s(n) represents the nth input symbol of the first step of the encoder; h1′ represents the linear feedback tap coefficient after pruning, h2′ represents the nonlinear feedback tap coefficient after pruning, mod represents the modular operation, and M is the number of levels of the pulse amplitude modulation signal.

[0068] As a preferred implementation, S2 includes: pruning the linear feedback tap coefficients and nonlinear feedback tap coefficients of the second feedforward-decision feedback equalizer based on the absolute value kernel, selecting the linear feedback tap coefficients whose absolute values ​​are greater than the first threshold and the nonlinear feedback tap coefficients whose absolute values ​​are greater than the second threshold to construct Tomlinson-Harasima precoding based on the absolute value kernel.

[0069] The pruning operation is expressed as:

[0070]

[0071] Where T1 is the first threshold, T2 为 The second threshold. The product terms h′1(k)x(nk) and h′2(k)|x(nk)+x(nq)| of the tap coefficients h′1 and h′2 after the pruning operation are assigned 0 and are directly discarded in the absolute value kernel-based Tomlinson-Halassima precoding and do not participate in the coding calculation. The corresponding constructed absolute value kernel-based Tomlinson-Halassima precoding can be expressed as:

[0072]

[0073] x(n)=mod 2M (x'(n)).

[0074] As a preferred implementation, before S1, it also includes: using the signal transmitting end to generate an original pulse amplitude modulation signal sequence, and up-sampling and pulse shaping the original pulse amplitude modulation signal sequence to obtain an initial pulse amplitude modulation signal sequence; modulating the initial pulse amplitude modulation signal sequence onto an optical carrier through intensity modulation and transmitting it through a transmission link; using the signal receiving end to convert the received optical signal into an electrical signal, and then down-sampling, clock recovery and matched filtering are performed to obtain a target pulse amplitude modulation signal.

[0075] Specifically, the dispersion and nonlinearity compensation method of Tomlinson-Halassima precoding based on absolute value kernel is divided into two stages. The first stage is the pre-training stage: the feedforward-decision feedback equalizer based on absolute value kernel is trained by using the pulse amplitude modulation received signal through the optical channel, and the tap coefficients of the trained decision feedback equalizer part are pruned as the feedback loop tap coefficients of Tomlinson-Halassima precoding based on absolute value kernel; the second stage is the transmission stage: the constructed Tomlinson-Halassima precoding based on absolute value kernel is used to process the transmitted pulse amplitude modulation data of the optical channel to realize the pre-compensation of the dispersion and nonlinear damage of the optical channel. The specific steps are as follows:

[0076] Step 1: In the pre-training stage, the signal transmitter generates a pulse amplitude modulation signal training sequence; upsamples and pulse shapes the pulse amplitude modulation signal; specifically, the binary training sequence is mapped to generate a 40Gbaud rate 4-level pulse amplitude modulation sequence; the 40Gbaud 4-level pulse amplitude modulation sequence is upsampled by 2 times and then pulse shaped, using a 0.6 roll-off factor root raised cosine filter; and resampled to the sampling rate of 60GSa / s required by the transmitter.

[0077] Step 2: The processed pulse amplitude modulation signal is modulated onto a C-band optical carrier through intensity modulation and transmitted through a standard single-mode optical fiber link. The optical intensity signal is converted into an electrical signal by sampling and direct reception at the signal receiving end. The intensity modulation / direct detection transmission system includes a signal transmitting end and a signal receiving end, such as Figure 3 , which is also configured with: an intensity modulation unit using a Mach Zehnder modulator (MZM) to modulate electrical signals onto intensity changes of an optical carrier. The laser wavelength is 1550nm, and a Mach Zehnder modulator is used to achieve optical intensity modulation; a standard single-mode optical fiber link is used to transmit optical signals over long distances, using a G.652 standard single-mode optical fiber with a length of 70km; an optical amplifier unit using an erbium-doped fiber amplifier (EDFA) to compensate for the power loss of optical signals transmitted over long distances; a direct detection unit for receiving optical signals and converting them into electrical signals, using a 40GHz bandwidth photodiode device, and using an 80GSa / s ADC for sampling.

[0078] Step 3: Downsample, clock recover and match filter the received signal at the receiving end; Downsample, clock recover and match filter the received signal at the receiving end; the downsampling process reduces the sampling rate of the received signal to 2 times the baud rate, and the clock recovery corrects the sampling clock deviation using the Gardner algorithm, and the matched filter is implemented using a root raised cosine filter with a roll-off factor of 0.6.

[0079] Step 4: Use the signal to train the tap coefficients of the feedforward-decision feedback equalizer based on the absolute value kernel; perform threshold pruning on the tap coefficients (h1) of the linear part and the tap coefficients (h2) of the nonlinear part of the obtained decision feedback equalizer part and use them as the Tomlinson-Halassima precoding tap coefficients based on the absolute value kernel.

[0080] Step 5: In the transmission phase, the data at the signal transmitter is mapped into a pulse amplitude modulation signal; dispersion and nonlinear impairments are pre-compensated through Tomlinson-Halassima precoding based on an absolute value kernel.

[0081] Example 2

[0082] This embodiment provides a communication method for data center optical interconnection, such as Figure 4 and Figure 5 As shown, the process includes the following steps.

[0083] A1: At the signal transmitting end, the dispersion and nonlinear compensation method for data center optical interconnection is used to compensate the pulse amplitude modulation signal to be transmitted, and modulate it to the optical carrier for transmission. For example, the data at the signal transmitting end is mapped to a pulse amplitude modulation signal; the 4-level pulse amplitude modulation signal is pre-compensated for dispersion and nonlinear damage by the absolute value kernel-based Tomlinson-Halassima precoding constructed in step S4; the signal after the absolute value kernel-based Tomlinson-Halassima precoding is upsampled by 2 times and pulse shaped, and the pulse shaping is based on a root raised cosine filter with a roll-off factor of 0.6; the signal is resampled to the sampling rate of 60GSa / s required by the transmitter.

[0084] A2: At the signal receiving end, the received optical signal is converted into the initial electrical signal, and then down-sampled, clock recovered and matched filtered to obtain the target electrical signal, and then the target electrical signal is decoded and restored to obtain the original pulse amplitude modulation signal. For example, the received signal is down-sampled, clock recovered and matched filtered at the receiving end; the down-sampling process reduces the sampling rate of the received signal to 2 times the baud rate, and the clock recovery corrects the sampling clock deviation, which is implemented using the Gardner algorithm, and the matched filter is implemented using a root raised cosine filter with a roll-off factor of 0.6.

[0085] As a preferred implementation, A2 includes: at the signal receiving end: converting the received optical signal into an initial electrical signal, and then performing downsampling, clock recovery and matched filtering to obtain a target electrical signal; using a third feedforward equalizer based on an absolute value kernel and a modular operation to perform Tomlinson-Harasima precoding decoding based on an absolute value kernel on the target electrical signal to restore the original pulse amplitude modulation signal; and demapping the pulse amplitude modulation signal;

[0086] The third feedforward equalizer based on the absolute value kernel is obtained by taking the original pulse amplitude modulation signal s(n) and the first and second step output sequences x′(n) and x(n) after the Tomlinson-Harasima precoding based on the absolute value kernel, and the result of the s(n)+x′(n)-x(n) operation as the reference signal, and taking the target pulse amplitude modulation signal sequence restored by the signal receiving end as the input signal, and adjusting the tap coefficients of the second feedforward equalizer based on the absolute value kernel through an adaptive algorithm for training; as a preferred implementation, the third feedforward equalizer based on the absolute value kernel is expressed as:

[0087]

[0088] Wherein, k, q both represent the ordinal number of the input signal, y(n) represents the nth output signal of the third feedforward equalizer based on the absolute value kernel, w1(k) represents the linear feedforward tap coefficient of the third feedforward equalizer when the kth input signal is input, w2(k,q) represents the nonlinear feedforward tap coefficient of the third feedforward equalizer when the kth input signal and the qth input signal are input, r(·) represents the equalizer input signal sequence, N1 represents the linear feedforward memory length, N2 represents the nonlinear feedforward memory length, Indicates a floor operation.

[0089] The Tomlinson-Harasima precoder decoder based on absolute value kernel uses a feedforward equalizer based on absolute value kernel to perform channel equalization on the signal, and then performs modulo 8 operation on the signal to restore the original 4-level pulse amplitude modulation signal; finally, the 4-level pulse amplitude modulation signal is demapped. The feedforward equalizer based on absolute value kernel can be expressed as:

[0090]

[0091] The present embodiment provides a method for improving the performance of an optical interconnection transmission system between data centers, which can effectively compensate for optical fiber dispersion damage and nonlinear damage caused by optoelectronic devices.

[0092] Compared with the prior art, this application has the following beneficial effects:

[0093] In the C-band intensity modulation / direct detection system, the direct detection signal after transmission is expressed as:

[0094]

[0095] c(t) is the DC bias of the signal, and s(t) is the original signal. -1 is the inverse Fourier transform. h(t) and H(f) are the dispersion-related transfer functions:

[0096] Re(H(f))=cos(2π 2β2Lf 2 );

[0097] β2 is the group velocity dispersion, and L is the fiber length. It is the signal-signal beat frequency interference term, which will introduce dispersion-related nonlinear distortion to the signal. In addition, the modulator, amplifier and other devices have nonlinear responses that cause signal distortion. Therefore, the nonlinear algorithm based on the absolute value kernel function enhances the nonlinear damage compensation capability compared to the traditional algorithm, and can better compensate for the dispersion channel response and device nonlinear damage.

[0098] The absolute value is used as the nonlinear kernel function in the present invention, which reduces the number of multipliers by half compared with the traditional nonlinear kernel based on Volterra series, and the threshold pruning step is introduced to further reduce the number of multipliers required. Therefore, the method of the present invention can compensate for the dispersion and nonlinear damage suffered by the C-band system during medium and long-distance transmission with low computational complexity, improve the quality of the received signal, and is suitable for application scenarios such as data center optical interconnection that is sensitive to cost and power consumption.

[0099] Figure 6 The bit error rate performance of the absolute value kernel-based Tomlinson-Halassima precoding method of the present invention is compared with that of the traditional linear Tomlinson-Halassima precoding method, the feedforward-decision feedback equalizer post-compensation method, and the Volterra-based Tomlinson-Halassima precoding method.

[0100] The mathematical models of the three solutions used for comparison are as follows:

[0101] 1. Linear Tomlinson-Halassima precoding can be expressed as:

[0102]

[0103] x(n)=mod 2M (x'(n))

[0104] The channel equalization of the decoder can be expressed as:

[0105] 2. Tomlinson-Halassima precoding based on Volterra can be expressed as:

[0106]

[0107] x(n)=mod 2M (x'(n))

[0108] The channel equalization of the decoder can be expressed as:

[0109]

[0110] 3. The feedforward-decision feedback equalizer is expressed as:

[0111]

[0112] The above-mentioned traditional V Tomlinson-Halassima precoding method and linear Tomlinson-Halassima precoding method differ only in the Tomlinson-Halassima precoder construction formula of step S3 of the signal processing at the transmitting end and the decoder construction formula of step S8. The traditional methods do not have the coefficient pruning process of step S4. The traditional feedforward-decision feedback equalizer post-compensation method has no pre-compensation at the transmitting end, and only uses the feedforward-decision feedback equalizer channel equalization to compensate for signal damage at the receiving end. The rest of the signal processing parts of these four schemes in the experiment are the same.

[0113] Figure 5 The relationship between the bit error rate and the received optical power using the above four schemes in a 70km 40Gbaud pulse amplitude modulation-4 intensity modulation / direct detection transmission experiment is shown. The received optical power is defined as Figure 3 The optical power before entering the EDFA module is controlled by an adjustable optical attenuator. The parameters of the above four dispersion and nonlinear compensation methods are shown in Table 1.

[0114] Table 1

[0115]

[0116] The absolute value kernel-based Tomlinson-Harasima precoding method of the present invention achieves 9.08×10 -3 , which is lower than the soft decision forward error correction threshold of 1×10 -2 . The performance of the Tomlinson-Halassima precoding based on the absolute value kernel is almost the same as that of the Tomlinson-Halassima precoding method based on Volterra, and is significantly higher than the feedforward-decision feedback equalizer post-compensation scheme. Compared with the linear Tomlinson-Halassima precoding method, the Tomlinson-Halassima precoding method based on the absolute value kernel of the present invention significantly improves the performance of the Tomlinson-Halassima precoding by adding an absolute value term nonlinear kernel in the transfer function. Table 2 shows the number of multipliers required per symbol for the dispersion and nonlinear compensation algorithm parts of several schemes. Compared with the Tomlinson-Halassima precoding method based on Volterra, the present invention reduces the number of multipliers by approximately 90% in the transmitter encoder and the receiver decoder parts while maintaining almost the same performance.

[0117] Table 2

[0118]

[0119] Example 3

[0120] This embodiment provides a data center optical interconnection system, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0121] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A dispersion and nonlinear compensation method for data center optical interconnection, characterized in that: include: S1: Taking the initial pulse amplitude modulation signal sequence sent by the signal transmitting end as the reference signal and the target pulse amplitude modulation signal sequence recovered by the signal receiving end as the input signal, the first feedforward-decision feedback equalizer based on the absolute value kernel is trained by adjusting the tap coefficient of the equalizer through an adaptive algorithm to obtain the second feedforward-decision feedback equalizer based on the absolute value kernel with the best equalization performance under the transmission link between the signal transmitting end and the signal receiving end; S2: pruning tap coefficients of a decision feedback equalizer in the second feedforward-decision feedback equalizer based on the absolute value kernel, and constructing a Tomlinson-Harasima precoding based on the absolute value kernel by using the pruned tap coefficients; S3: Compensate the pulse amplitude modulation signal to be transmitted by using the absolute value kernel-based Tomlinson-Harasima precoding to eliminate the dispersion and nonlinear damage caused by the transmission link to the transmission signal.

2. The dispersion and nonlinear compensation method for data center optical interconnection according to claim 1, characterized in that: The S2 includes: The nonlinear feedback tap coefficients of the decision feedback equalizer in the second feedforward-decision feedback equalizer based on the absolute value kernel are pruned, and the nonlinear feedback tap coefficients greater than a threshold are selected to construct Tomlinson-Harasima precoding based on the absolute value kernel.

3. The dispersion and nonlinear compensation method for data center optical interconnection according to claim 2, characterized in that: The second feedforward-decision feedback equalizer based on the absolute value kernel includes a second feedforward equalizer and a second decision feedback equalizer. The second feedforward-decision feedback equalizer based on the absolute value kernel is expressed as: Wherein, n is the ordinal number of the output signal, k and q both represent the ordinal numbers of the input signals, y(n) represents the nth output signal of the second feedforward-decision feedback equalizer based on the absolute value kernel, w1(k) represents the linear feedforward tap coefficient of the second feedforward equalizer when the kth input signal is input, w2(k,q) represents the nonlinear feedforward tap coefficient of the second feedforward equalizer when the kth input signal and the qth input signal are input, r(·) represents the equalizer input signal sequence, h1(k) represents the linear feedback tap coefficient of the second decision feedback equalizer, h2(k,q) represents the nonlinear feedback tap coefficient of the second decision feedback equalizer, d(·) represents the decision symbol sequence corresponding to the output signal sequence y(n); N1 represents the linear feedforward memory length, N2 represents the nonlinear feedforward memory length, D1 represents the linear feedback memory length, and D2 represents the nonlinear feedback memory length. Indicates a floor operation.

4. The dispersion and nonlinear compensation method for data center optical interconnection according to claim 3, characterized in that: The absolute value kernel-based Tomlinson-Halassima precoding constructed in S2 is expressed as: x(n)=mod 2M (x'(n)); Among them, x(n), x(nk) and x(nq) represent the nth, nkth and nqth output signals of the second step output sequence of the encoder, x′(n) is the nth output signal of the first step output sequence of the encoder, s(n) represents the nth input symbol of the first step of the encoder; h′2 represents the nonlinear feedback tap coefficient after pruning, mod represents modular operation, and M is the number of levels of the pulse amplitude modulation signal.

5. The dispersion and nonlinear compensation method for data center optical interconnection according to claim 1, characterized in that: The S2 includes: The linear feedback tap coefficients and the nonlinear feedback tap coefficients of the second feedforward-decision feedback equalizer based on the absolute value kernel are pruned, and the linear feedback tap coefficients whose absolute values ​​are greater than a first threshold and the nonlinear feedback tap coefficients whose absolute values ​​are greater than a second threshold are selected to construct Tomlinson-Harasima precoding based on the absolute value kernel.

6. The dispersion and nonlinear compensation method for data center optical interconnection according to claim 1, characterized in that: Before S1, it also includes: Generate an original pulse amplitude modulation signal sequence by using the signal transmitting end, and perform up-sampling and pulse shaping on the original pulse amplitude modulation signal sequence to obtain the initial pulse amplitude modulation signal sequence; modulate the initial pulse amplitude modulation signal sequence onto an optical carrier through intensity modulation and transmit it via the transmission link; The signal receiving end is used to convert the received optical signal into an electrical signal, and then down-sampling, clock recovery and matched filtering are performed to obtain the target pulse amplitude modulation signal.

7. A communication method for optical interconnection in a data center, characterized in that: include: A1: At the signal transmitting end, the pulse amplitude modulation signal to be transmitted is compensated by using the dispersion and nonlinear compensation method for data center optical interconnection as described in any one of claims 1 to 6, and modulated to an optical carrier for transmission; A2: At the signal receiving end, the received optical signal is converted into an initial electrical signal, and then down-sampled, clock recovered and matched filtered to obtain a target electrical signal, and the target electrical signal is decoded to restore the original pulse amplitude modulation signal.

8. The communication method for data center optical interconnection according to claim 7, characterized in that: The step A2 of decoding and restoring the target electrical signal to obtain the original pulse amplitude modulation signal includes: Using a third feedforward equalizer based on an absolute value kernel and a modular operation to perform Tomlinson-Harasima precoding decoding based on an absolute value kernel on the target electrical signal to restore the original pulse amplitude modulation signal; Among them, the third feedforward equalizer based on absolute value kernel is obtained by taking the original pulse amplitude modulation signal s(n) and the first and second step output sequences x′(n) and x(n) after Tomlinson-Harasima precoding based on absolute value kernel, and the result of s(n)+x′(n)-x(n) operation as reference signals, and taking the target pulse amplitude modulation signal sequence recovered by the signal receiving end as input signal, and adjusting the tap coefficients of the second feedforward equalizer based on absolute value kernel through adaptive algorithm for training.

9. The communication method for data center optical interconnection according to claim 8, characterized in that: The third feedforward equalizer based on absolute value kernel is expressed as: Wherein, k and q both represent the ordinal number of the input signal, y(n) represents the nth output signal of the third feedforward equalizer based on the absolute value kernel, w1(k) represents the linear feedforward tap coefficient of the third feedforward equalizer when the kth input signal is input, w2(k,q) represents the nonlinear feedforward tap coefficient of the third feedforward equalizer when the kth input signal and the qth input signal are input, r(·) represents the equalizer input signal sequence, N1 represents the linear feedforward memory length, N2 represents the nonlinear feedforward memory length, Indicates a floor operation.

10. A data center optical interconnection communication system, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 7 to 9 are implemented.

Images