A maximum likelihood sequence estimation method and system with reduced state number and transition path number

Abstract

The application relates to the technical field of optical communication systems, and discloses a maximum likelihood sequence estimation method and system with reduced state number and transition path number, which comprises the following steps: S1. receiving a signal; performing equalization on the received signal through an LMS equalizer to obtain an equalized signal; S2. inputting the equalized signal into a Threshold Detector to perform reduced state pre-decision, obtaining TD_Seq; inputting the equalized signal into a Slicer to perform reduced transition path pre-decision, obtaining S_Seq; inputting the equalized signal into a Postfilter to filter high-frequency noise, obtaining PF_Seq; S3. inputting TD_Seq, S_Seq and PF_Seq into RST-MLSE respectively to perform processing and construct a grid graph of RST-MLSE; and S4. performing backtracking through the grid graph to obtain an estimated maximum likelihood probability of a transmitted signal sequence. The application solves the problems of existing technology, such as complex calculation, limited bandwidth and code inter-symbol interference caused by fiber dispersion, and has the characteristic of not depending on a Slicer parameter W.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of optical communication systems, and more particularly, to a maximum likelihood sequence estimation method and system with reduced state number and transition path number. BACKGROUND

[0002] With the booming development of 4K / 8K video, cloud services and 5G communication, the demand for large-capacity transmission of optical access networks and data center interconnections is rapidly increasing. Especially for the next generation 800GbE or 1.6TbE standard, the baud rate of the signal will increase rapidly. However, the increase of device bandwidth based on CMOS technology is insufficient, resulting in a gradually increasing gap between signal bandwidth and device bandwidth. Therefore, the inter-symbol interference (ISI) caused by bandwidth limitation has become an urgent problem to be solved. In addition, chromatic dispersion (CD) in optical fiber also distorts the time-domain waveform and exacerbates the ISI problem. Compared with traditional minimum mean square error equalizer and decision feedback equalizer, MLSE has shown superior ability in solving ISI problems. However, since the computational complexity of MLSE increases exponentially with the memory length (L), its actual deployment in IM / DD systems is problematic. Therefore, how to reduce the computational complexity of MLSE has been studied, and some methods have been proposed:

[0003] One method mainly uses the signal after LMS equalization, which is close to the standard PAM signal level to some extent, so a pre-decision is made on the equalized signal. Taking PAM-4 signal as an example, when the memory length is 2, the state in the MLSE grid is {-3,-1,1,3} without pre-decision, and after pre-decision, the state becomes {-3,-1} or {-1,1} or {1,3}. However, this method has limited complexity reduction, and the complexity reduction is constant for any received optical power, which cannot dynamically change and therefore has certain limitations.

[0004] One method mainly uses a lookup table corresponding to the MLSE memory signal established during equalization to replace the linear convolution process in MLSE calculation. However, this method only reduces the linear convolution process, and the complexity reduction is very limited.

[0005] DRP-MLSE algorithm based on decision region partitioning:

[0006] A method mainly uses a Slicer to filter out signals that can be accurately determined by hard decision, and the remaining signals are determined by RS-MLSE. The operation of the Slicer depends on a parameter W to control the range of the filtered signals. DRP-MLSE can only significantly reduce the complexity when the value of W is large. Therefore, when the bandwidth is severely limited, this method is not applicable and cannot significantly reduce the complexity.

[0007] The RS-MLSE algorithm can only reduce the number of states in MLSE, so it can only bring a fixed reduction in complexity and cannot dynamically change according to the quality of the received signal. The MLSE algorithm based on lookup table only uses a lookup table to replace the linear convolution process in MLSE, so the amount of calculation is limited and the complexity cannot be compressed. The DRP-MLSE algorithm depends on a large Slicer filter parameter W to significantly reduce the complexity. For bandwidth limitations, W cannot be too large when there is dispersion. Therefore, this method cannot significantly reduce the complexity.

[0008] The existing MLSE simplification detection method and device based on MMSE are applied in a single carrier system, wherein the method comprises: generating a preliminary sequence from a user signal according to a channel estimation value and a minimum mean square error MMSE criterion; performing peripheral extension on the preliminary sequence to form a sequence set to be determined; and traversing the set to select a sequence having the maximum likelihood with the user signal as the detection result according to a maximum likelihood criterion. The device is provided with an MMSE detector (31) before the ML detector (32) for preprocessing. The method and device greatly reduce the complexity of the MLSE algorithm, and the performance is better than that of the MMSE algorithm. On the basis of ensuring a certain performance gain, the complexity is greatly reduced at the cost of a small amount of performance, so that the simplified detection method and device have general requirements for hardware and processing capacity, thereby facilitating practical application.

[0009] However, the prior art still has the problems of calculation complexity, bandwidth limitation and inter-symbol interference caused by fiber dispersion. Therefore, how to invent a low-complexity MLSE algorithm is a problem that needs to be solved in the technical field. SUMMARY

[0010] The present application provides a low-complexity MLSE algorithm to solve the problems of calculation complexity, bandwidth limitation and inter-symbol interference caused by fiber dispersion in the prior art, which has the characteristic of not depending on the Slicer parameter W.

[0011] To achieve the above-mentioned purposes of the present application, the technical solutions adopted are as follows:

[0012] A maximum likelihood sequence estimation method for reducing state number and transition path number, comprising the following steps:

[0013] S1. Receiving a signal; equalizing the received signal through an LMS equalizer to obtain an equalized signal;

[0014] S2. Inputting the equalized signal into a Threshold Detector for reduced state pre-decision to obtain TD_Seq; inputting the equalized signal into a Slicer for reduced transition path pre-decision to obtain S_Seq; and inputting the equalized signal into a Postfilter to filter out high-frequency noise to obtain PF_Seq;

[0015] S3. Inputting TD_Seq, S_Seq and PF_Seq into a RST-MLSE respectively for processing to construct a grid chart of the RST-MLSE;

[0016] S4. Backtracking through the grid chart to obtain an estimated maximum likelihood probability of a transmitted signal sequence.

[0017] Preferably, in step S1, the equalized signal is arranged according to time, and the equalized signal at time k is denoted as LMS_Eq_Seq(k).

[0018] Further, in step S2, the equalized signal at the current time is pre-decided by the Threshold Detector, and the specific steps are as follows:

[0019] S201. Inputting the equalized signal at the current time into the Threshold Detector and comparing it with the threshold value of the Threshold Detector;

[0020] S202. Judging the Region of the Threshold Detector of the equalized signal at the current time according to the size of the equalized signal at the current time;

[0021] S203. Obtaining TD_Seq according to the region number to which the equalized signal at the current time belongs.

[0022] The Threshold Detector is provided with three regions, and LMS_Eq_Seq(k) is , wherein region 1 is , region 2 is , and region 3 is .

[0023] Further, in step S2, the Slicer is used to make a transition path prediction for the equalized signal at the current time to obtain S_Seq, specifically as follows:

[0024] S2201. The equalized signal at the current time is input into the Slicer for comparison to determine the shadow region of the Slicer in which the equalized signal at the current time is located.

[0025] S2202. The information of the shadow region in which the equalized signal at the current time is located is recorded to obtain S_Seq.

[0026] Further, the Slicer has four shadow regions.

[0027] Further, in step S2201, the shadow region of the Slicer in which the equalized signal at the current time is located is determined, specifically as follows:

[0028] wherein, L(k) represents the information of the shadow region in which the signal at the current time k is located, LMS_Eq_Seq(k) represents the equalized signal at the current time, Null represents a null value, Slicer represents a filter parameter of the Slicer, which is used to adjust the range of the shadow region.

[0029] Further, in step S2, after the Threshold Detector makes a state prediction for the equalized signal at the current time, if the equalized signal at the current time belongs to Region 1, then the state of the MLSE trellis graph at the current time only considers {-3, -1}; if the equalized signal at the current time belongs to Region 2, then the state of the MLSE trellis graph at the current time only considers {-1, 1}; if the equalized signal at the current time belongs to Region 3, then the state of the MLSE trellis graph at the current time only considers {1, 3}.

[0030] Further, in the step S2, after the Slicer performs the reduced transition path pre-decision on the equalized signal at the current time, for the time n, if S_Seq(n-1)= null and S_Seq(n)= null, no additional transition path can be compressed in the MLSE grid; if S_Seq(n-1)≠ null and S_Seq(n)= null, the two state transitions at the time n are both changed from the most probable state recorded at the time n-1; the most probable state is the standard PAM-4 symbol contained in the shadow area; if S_Seq(n-1)= null and S_Seq(n)≠ null, only the most probable state recorded at the time n is subjected to the transition path measurement calculation and the survivor path selection, and for the other state at the time n, the accumulated measurement is set to infinity, indicating that the state cannot be reached; if S_Seq(n-1)≠ null and S_Seq(n)≠ null, no transition path measurement calculation and comparison are performed when the MLSE grid is constructed, and only the state transition is recorded.

[0031] A maximum likelihood sequence estimation system with reduced state number and transition path number, comprising a signal receiving module, an LMS equalizer module, a Threshold Detector module, a Slicer module, a Postfilter module, an RST-MLSE module and a backtracking module.

[0032] The signal receiving module is used for receiving a signal.

[0033] The LMS equalizer module is used for equalizing the received signal.

[0034] The Threshold Detector module is used for performing reduced state pre-decision on the equalized signal to obtain TD_Seq.

[0035] The Slicer module is used for performing reduced transition path pre-decision on the equalized signal to obtain S_Seq.

[0036] The Postfilter module inputs the equalized signal into the Postfilter to filter out high-frequency noise to obtain PF_Seq.

[0037] The RST-MLSE module is used for constructing the grid graph of the RST-MLSE according to TD_Seq, S_Seq and PF_Seq.

[0038] The backtracking module is used for backtracking through the grid graph to obtain an estimated maximum likelihood probability of the transmitted signal sequence.

[0039] The present application has the following advantages:

[0040] The application provides a low-complexity maximum likelihood sequence estimation method for reducing the number of states and the number of transition paths, so as to realize similar bit error rate performance as a traditional MLSE algorithm and dynamically reduce the MLSE calculation complexity under different received optical powers; specifically, the application inputs the signal equalized by an LMS equalizer into a Threshold Detector for state reduction pre-decision to obtain TD_Seq, inputs the signal into a Slicer for transition path reduction pre-decision, and inputs the signal into a Postfilter to filter high-frequency noise to obtain PF_Seq; finally, the application inputs TD_Seq, S_Seq and PF_Seq into RST-MLSE for processing, constructs a grid graph of RST-MLSE, and backtracks through the grid graph to obtain an estimated maximum likelihood probability of the transmitted signal sequence. Therefore, the application solves the problems of calculation complexity, bandwidth limitation and inter-symbol interference caused by fiber dispersion in the prior art, and provides a low-complexity MLSE algorithm which has the characteristic of not depending on the Slicer parameter W. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 It is a flow chart of the maximum likelihood sequence estimation method for reducing the number of states and the number of transition paths.

[0042] Figure 2 It is a flow chart of the maximum likelihood sequence estimation method for reducing the number of states and the number of transition paths.

[0043] Figure 3 It is a running mechanism schematic diagram of the Threshold Detector of the maximum likelihood sequence estimation method for reducing the number of states and the number of transition paths.

[0044] Figure 4 It is a running mechanism schematic diagram of the Slicer algorithm of the maximum likelihood sequence estimation method for reducing the number of states and the number of transition paths.

[0045] Figure 5 It is a mode schematic diagram of the maximum likelihood sequence estimation method for reducing the number of states and the number of transition paths.

[0046] Figure 6 It is a grid graph schematic diagram of the maximum likelihood sequence estimation method for reducing the number of states and the number of transition paths.

[0047] Figure 7 It is an experimental setting diagram of the maximum likelihood sequence estimation method for reducing the number of states and the number of transition paths.

[0048] Figure 8is a bit error rate comparison chart of a maximum likelihood sequence estimation method of reducing state number and transition path number under 125Gbit / s PAM-4 signal BTB transmission.

[0049] Figure 9 is a MLSE calculation complexity reduction percentage comparison chart of a maximum likelihood sequence estimation method of reducing state number and transition path number under 125Gbit / s PAM-4 signal 2 km SSMF transmission.

[0050] Figure 10 is a bit error rate comparison chart of a maximum likelihood sequence estimation method of reducing state number and transition path number under 125Gbit / s PAM-4 signal 2 km SSMF transmission.

[0051] Figure 11 is a MLSE calculation complexity reduction percentage comparison chart of a maximum likelihood sequence estimation method of reducing state number and transition path number under 125Gbit / s PAM-4 signal 2 km SSMF transmission.

[0052] Figure 12 is a bit error rate comparison chart of a maximum likelihood sequence estimation method of reducing state number and transition path number under 130Gbit / s PAM-4 signal BTB transmission.

[0053] Figure 13 is a MLSE calculation complexity reduction percentage comparison chart of a maximum likelihood sequence estimation method of reducing state number and transition path number under 130Gbit / s PAM-4 signal BTB transmission.

[0054] Figure 14 is a bit error rate comparison chart of a maximum likelihood sequence estimation method of reducing state number and transition path number under 130Gbit / s PAM-4 signal 2 km SSMF transmission.

[0055] Figure 15 is a MLSE calculation complexity reduction percentage comparison chart of a maximum likelihood sequence estimation method of reducing state number and transition path number under 130Gbit / s PAM-4 signal 2 km SSMF transmission.

[0056] Figure 16 is a system principle diagram of a maximum likelihood sequence estimation system of reducing state number and transition path number. DETAILED DESCRIPTION

[0057] The application will be described in detail below in conjunction with the drawings and specific embodiments.

[0058] Embodiment 1

[0059] As Figure 1 shown in FIG. 1, a maximum likelihood sequence estimation method for reducing the number of states and the number of transition paths comprises the following steps:

[0060] S1. receiving a signal; equalizing the received signal through an LMS equalizer to obtain an equalized signal;

[0061] S2. inputting the equalized signal into a Threshold Detector for reduced state pre-decision to obtain TD_Seq; inputting the equalized signal into a Slicer for reduced transition path pre-decision to obtain S_Seq; inputting the equalized signal into a Postfilter to filter out high frequency noise to obtain PF_Seq;

[0062] S3. inputting TD_Seq, S_Seq, and PF_Seq into an RST-MLSE for processing to construct a trellis diagram of the RST-MLSE;

[0063] S4. backtracking through the trellis diagram to obtain an estimated maximum likelihood probability of the transmitted signal sequence.

[0064] Embodiment 2

[0065] As Figure 2 shown in FIG. 1, a maximum likelihood sequence estimation method for reducing the number of states and the number of transition paths comprises the following steps:

[0066] S1. receiving a signal; equalizing the received signal through an LMS equalizer to obtain an equalized signal;

[0067] S2. inputting the equalized signal into a Threshold Detector for reduced state pre-decision to obtain TD_Seq; inputting the equalized signal into a Slicer for reduced transition path pre-decision to obtain S_Seq; inputting the equalized signal into a Postfilter to filter out high frequency noise to obtain PF_Seq;

[0068] S3. inputting TD_Seq, S_Seq, and PF_Seq into an RST-MLSE for processing to construct a trellis diagram of the RST-MLSE;

[0069] S4. backtracking through the trellis diagram to obtain an estimated maximum likelihood probability of the transmitted signal sequence.

[0070] In this embodiment, the 0-1 bitstream sequence is first mapped to a standard PAM-4 signal in the DSP at the transmitting end. Then, the signal is Nyquist shaped using a root-raised cosine filter with a roll-off factor of 0.1. After that, the shaped signal is downsampled and then input into a digital-to-analog converter with a bandwidth of 16 GHz and a sampling rate of 80 GSa / s.

[0071] In this embodiment, in the DSP at the receiving end, the signal obtained by PD detection is mainly subjected to upsampling, matched filtering, synchronization, LMS equalization, Threshold Detector, Slicer, Postfilter, RST-MLSE, PAM-4 signal demapping and bit error rate calculation.

[0072] like Figure 7 As shown, the experimental system used in this embodiment is a bandwidth-limited IM / DD system. First, at the transmitting end, offline data generated by MATLAB is loaded onto a DAC with an 80 GSa / s sampling rate and a 3 dB bandwidth of 16 GHz for digital-to-analog conversion. Then, an electrical attenuator is used to mitigate nonlinear distortion, followed by signal amplification using an electrical amplifier with a gain of 23 dB. The transmitted signal after electrical amplification, a 1550 nm laser, and a DC bias are fed into a Mach-Zehnder modulator to achieve electro-optical conversion. The optical carrier of the modulator comes from a tunable external cavity laser. The optical signal is then transmitted through a BTB or a 2 km standard single-mode fiber. After transmission, a variable optical attenuator is used to adjust the received optical power of the received signal. The attenuated optical signal is then converted to photoelectric signal using a PD. The analog electrical signal output by the PD is acquired by an oscilloscope with a cutoff bandwidth of 36 GHz and a sampling rate of 80 GSa / s for offline DSP.

[0073] In this embodiment, the frequency response of the system was measured using training symbols before the experiment began. The bandwidths for -3 dB and -10 dB were 2.4 GHz and 16.2 GHz, respectively, showing that the system's bandwidth is severely limited.

[0074] In one specific embodiment, in step S1, the equalized signals are arranged according to time, and the equalized signal at time k is represented as LMS_Eq_Seq(k).

[0075] In one specific embodiment, step S2 involves using a Threshold Detector to perform a state reduction pre-decision on the equalized signal at the current moment. The specific steps are as follows:

[0076] S201. input the equalized signal at the current time into the Threshold Detector, and compare the equalized signal with the threshold value of the Threshold Detector;

[0077] S202. determine the Region of the Threshold Detector to which the equalized signal at the current time belongs according to the size of the equalized signal at the current time;

[0078] S203. obtain TD_Seq according to the Region number to which the equalized signal at the current time belongs.

[0079] As shown in the figure, the Threshold Detector has three Regions, and LMS_Eq_Seq(k) is Figure 3 , wherein Region 1 is , Region 2 is , and Region 3 is . .

[0080] In one embodiment, in step S2, the equalized signal at the current time is subjected to a reduced transition path pre-decision by the Slicer to obtain S_Seq, and the specific process is as follows:

[0081] S2201. input the equalized signal at the current time into the Slicer to compare and determine the shadow Region of the Slicer to which the equalized signal at the current time belongs;

[0082] S2202. record the information of the shadow Region to which the equalized signal at the current time belongs to obtain S_Seq.

[0083] In one embodiment, the Slicer has four shadow Regions.

[0084] As shown in the figure, in one embodiment, in step S2201, the shadow Region of the Slicer to which the equalized signal at the current time belongs is determined, and the specific process is as follows: Figure 4

[0085] wherein represents the information of the shadow Region to which the signal at the current time k belongs, i.e. LMS_Eq_Seq(k), represents the equalized signal at the current time, Null represents null value, represents the Slicer filtering parameter, which is used to adjust the range of the shadow Region.

[0086] ​In one specific embodiment, in step S2, after the Threshold Detector makes a reduced state decision on the equalized signal at the current time, if the equalized signal at the current time belongs to Region 1, then the state at the current time in the MLSE trellis constructed only considers {-3, -1}; if the equalized signal at the current time belongs to Region 2, then the state at the current time in the MLSE trellis constructed only considers {-1, 1}; if the equalized signal at the current time belongs to Region 3, then the state at the current time in the MLSE trellis constructed only considers {1, 3}.

[0087] As shown in Figure 5 In one specific embodiment, in step S2, after the Slicer makes a reduced transition path decision on the equalized signal at the current time, for time n, if S_Seq(n-1) = null and S_Seq(n) = null, then the MLSE trellis constructed cannot compress any transition path additionally; if S_Seq(n-1) ≠ null and S_Seq(n) = null, then the two state transitions at time n are both from the most probable state recorded at time n-1; the most probable state is the standard PAM-4 symbol contained in the shadow region; if S_Seq(n-1) = null and S_Seq(n) ≠ null, then only the most probable state recorded at time n is subjected to transition path metric calculation and survivor path selection, and for the other state at time n, the accumulated metric is set to infinity, indicating that this state cannot be reached; if S_Seq(n-1) ≠ null and S_Seq(n) ≠ null, then no transition path metric calculation and comparison is performed when the MLSE trellis is constructed, and only the state transition is recorded.

[0088] In this embodiment, as shown in Figure 6 (a) is the output segment of the LMS equalizer, the Threshold Detector, the Slicer, and the Postfilter under 130 Gbit / s PAM-4 signal BTB transmission; (b) is the trellis before the survivor path is selected at each time; and (c) is the trellis after the survivor path is selected at each time. Figure 6 The process of constructing the RST-MLSE trellis using a received signal sequence in the experiment is shown. In this embodiment, when the output of the Slicer at a certain time is not equal to null, then there is no MLSE backtracking process, and the most probable state recorded in S_Seq at this time is the output of the RST-MLSE.

[0089] The bandwidth of the experimental system in this embodiment is severely limited. High-speed signals will produce serious inter-symbol interference after passing through the system, and there will also be a dispersion problem when passing through the optical fiber, further aggravating the influence of inter-symbol interference. Using a common equalizer cannot reach the HD-FEC threshold, so the MLSE scheme needs to be used to solve this problem. In the experiment, high-speed PAM-4 signals of 125 Gbit / s and 130 Gbit / s are tested. Different MLSE simplification schemes are used in the offline DSP at the receiving end to compare the bit error rate and the reduced complexity. The values in the brackets after DRP-MLSE and RST-MLSE in the figure legend represent the value of W in the corresponding Slicer.

[0090] In this embodiment, as shown in Figure 8~Figure 11 The relationship between the bit error rate of a 5 Gbit / s PAM-4 signal under BTB and 2 km SSMF transmission and the percentage reduction in MLSE calculation complexity is also shown. Due to severe bandwidth limitations, the bit error rate after LMS equalization cannot reach the hard decision threshold. All MLSE schemes can significantly improve the bit error rate performance. The bit error rate performance of RS-MLSE is similar to that of traditional MLSE, and the bit error rate performance of RST-MLSE is similar to that of DRP-MLSE. Although there is a ~0.13 and ~0.14 dB loss in received optical power between traditional MLSE and RST-MLSE in BTB and 2 km SSMF transmission, the values are very small and can be ignored. It is worth noting that the reduced MLSE calculation complexity of RST-MLSE is the largest. RS-MLSE can only reduce the calculation complexity by 75% at each ROP, while using DRP-MLSE can only obtain a lower improvement, which shows that DRP-MLSE cannot effectively solve the problem of severe bandwidth limitation and dispersion. In 125 Gbit / s BTB and 2 km SSMF transmission, the RST-MLSE proposed in this application can further reduce the calculation complexity by 7.9% at ROP of -6 dBm, which is a significant improvement, further compressing the calculation complexity and meeting the requirements of low-cost IM / DD systems.

[0091] In this embodiment, in order to further test the performance of RST-MLSE, the bit rate is increased to 130 Gbit / s, resulting in more severe bandwidth limitations. The results of BTB and 2 km SSMF transmission are shown in Figure 12~Figure 15As shown in the figure. In BTB transmission, there is about 0.13 dB loss of received optical power between the traditional MLSE and RST-MLSE at the hard decision threshold, but RST-MLSE can reduce the MLSE calculation complexity by about 87.5% than the traditional MLSE. RST-MLSE still has obvious improvement compared with DRP-MLSE, and the MLSE calculation complexity is further reduced by about 9% when ROP is-6 dBm. The performance of RST-MLSE in 130 Gbit / s 2 km SSMF transmission is equivalent to BTB, and there is about 0.11 dB loss of received optical power and 6.16% MLSE calculation complexity is further reduced. The above results show that RST-MLSE has very good performance in dealing with bandwidth limitation and dispersion problems, and the calculation complexity is significantly reduced, and has good application prospect in low-cost data center optical interconnection.

[0092] In the embodiment, as shown in Table 1, the number of additions and multiplications required for different MLSE schemes to establish a part of the MLSE trellis graph at each time is shown. Since the four cases in the MLSE trellis graph of the application may occur, the average of the number of additions and multiplications of the application is calculated according to the occurrence probability of each case in the four cases as a representative. Figure 5 Figure 5 P1, P2, P3 respectively represent the probabilities of the occurrence of cases (1), (2), (3) in the four cases. Figure 5

[0093] Table 1

[0094]

[0095] The application provides a low-complexity maximum likelihood sequence estimation method for reducing the number of states and transition paths, which can achieve similar bit error rate performance as the traditional MLSE algorithm and dynamically reduce the MLSE calculation complexity at different received optical powers. Specifically, the application inputs the signal equalized by the LMS equalizer into the ThresholdDetector for reduced state pre-decision to obtain TD_Seq, inputs the signal into the Slicer for reduced transition path pre-decision to obtain S_Seq, and inputs the signal into the Postfilter to filter high-frequency noise to obtain PF_Seq. Finally, the application inputs TD_Seq, S_Seq and PF_Seq into RST-MLSE for processing, constructs the trellis graph of RST-MLSE, and obtains the estimated maximum likelihood probability of the transmitted signal sequence through backtracking of the trellis graph.

[0096] ​​In addition to the above innovations, the present application can also effectively reduce the complexity at a smaller W value. Using this algorithm, 125 Gbit / s and 130 Gbit / s PAM-4 signals are successfully transmitted in the C band, back-to-back and over 2 km of standard single-mode optical fiber. The statistical bit error rate after transmission is lower than the hard decision forward error correction threshold, i.e. 3.8x10 -3 and this algorithm achieves approximately the same bit error rate performance as the traditional MLSE algorithm without reducing complexity. The present application thus solves the problems of existing technology of high computational complexity, limited bandwidth and inter-symbol interference caused by fiber dispersion, and provides a low-complexity MLSE algorithm that does not depend on the Slicer parameter W.

[0097] Embodiment 3

[0098] As shown in the Figure 16 A maximum likelihood sequence estimation system with reduced state number and transition path number includes a signal receiving module, an LMS equalizer module, a Threshold Detector module, a Slicer module, a Postfilter module, an RST-MLSE module, and a backtracking module.

[0099] The signal receiving module is configured to receive a signal.

[0100] The LMS equalizer module is configured to equalize the received signal.

[0101] The Threshold Detector module is configured to perform a reduced state pre-decision on the equalized signal to obtain TD_Seq.

[0102] The Slicer module is configured to perform a reduced transition path pre-decision on the equalized signal to obtain S_Seq.

[0103] The Postfilter module inputs the equalized signal into a Postfilter to filter out high-frequency noise to obtain PF_Seq.

[0104] The RST-MLSE module is configured to construct a trellis diagram of the RST-MLSE according to TD_Seq, S_Seq, and PF_Seq.

[0105] The backtracking module is configured to backtrack through the trellis diagram to obtain an estimated maximum likelihood probability of the transmitted signal sequence.

[0106] Obviously, the above-mentioned embodiments of the present application are only examples for clearly illustrating the present application, and are not intended to limit the implementation modes of the present application. Any modification, equivalent replacement and improvement, etc. within the spirit and principle of the present application should be included in the protection scope of the claims of the present application.

Claims

1. A maximum likelihood sequence estimation method for reducing the number of states and the number of transition paths, characterized by: The method comprises the following steps: S1. receiving a signal; equalizing the received signal through an LMS equalizer to obtain an equalized signal; The equalized signal is arranged according to time, and the equalized signal at time k is denoted as LMS_Eq_Seq(k); S2. inputting the equalized signal into a Threshold Detector for reduced state pre-decision to obtain TD_Seq; inputting the equalized signal into a Slicer for reduced transition path pre-decision to obtain S_Seq; and inputting the equalized signal into a Postfilter to filter out high-frequency noise to obtain PF_Seq; The equalized signal at the current time is pre-decided by the Threshold Detector, and the specific steps are as follows: S201. inputting the equalized signal at the current time into the Threshold Detector and comparing it with the threshold value of the Threshold Detector; S202. determining the Region of the Threshold Detector to which the equalized signal at the current time belongs according to the size of the equalized signal at the current time; S203. obtaining TD_Seq according to the region number to which the equalized signal at the current time belongs; and inputting the equalized signal into a Postfilter to filter out high-frequency noise to obtain PF_Seq; The equalized signal at the current time is pre-decided by the Slicer to obtain S_Seq, and the specific steps are as follows: S2201. inputting the equalized signal at the current time into the Slicer to determine the shadow region of the Slicer in which the equalized signal at the current time is located; S2202. recording the information of the shadow region in which the equalized signal at the current time is located to obtain S_Seq; The Threshold Detector has 3 regions, let LMS_Eq_Seq(k) be where region 1 is , region 2 is , and region 3 is ; The Slicer has four shadow regions; In step S2201, the shadow region of the Slicer in which the equalized signal at the current time is located is determined, and the specific steps are as follows: wherein, represents information of the shadow region where the signal at the current time k is located, i.e. LMS_Eq_Seq(k), represents the signal after equalization at the current time, Null represents null, represents a Slicer filter parameter, used to adjust the range of the shadow region; In step S2, after the equalized signal at the current time is pre-decided by the Threshold Detector, if the equalized signal at the current time belongs to Region 1, only {-3, -1} is considered for the state of the current time in the MLSE grid graph constructed; if the equalized signal at the current time belongs to Region 2, only {-1, 1} is considered for the state of the current time in the MLSE grid graph constructed; and if the equalized signal at the current time belongs to Region 3, only {1, 3} is considered for the state of the current time in the MLSE grid graph constructed. After the Slicer makes the reduced state pre-decision on the equalized signal at the current time, for time n, if S_Seq(n-1)= null and S_Seq(n)= null, no additional transition path can be compressed in the MLSE grid; if S_Seq(n-1)≠ null and S_Seq(n)= null, both state transitions at time n are from the most probable state recorded at time n-1; the most probable state is the standard PAM-4 symbol contained in the shadow area; if S_Seq(n-1)= null and S_Seq(n)≠ null, only the most probable state recorded at time n is subjected to the transition path metric calculation and survivor path selection, and for the other state at time n, its accumulated metric is set to infinity, indicating that this state cannot be reached; if S_Seq(n-1)≠ null and S_Seq(n)≠ null, no transition path metric calculation and comparison is performed when the MLSE grid is constructed, and only the state transition is recorded; S3. input TD_Seq, S_Seq and PF_Seq into the RST-MLSE for processing, and construct the grid graph of the RST-MLSE; S4. backtracking is performed through the grid graph to obtain the estimated maximum likelihood probability of the transmitted signal sequence.

2. A maximum likelihood sequence estimation system with reduced state number and transition path number, comprising: The method for implementing the method of claim 1 comprises a signal receiving module, an LMS equalizer module, a Threshold Detector module, a Slicer module, a Postfilter module, an RST-MLSE module, and a backtracking module; The signal receiving module is used to receive a signal; The LMS equalizer module is used to equalize the received signal; The Threshold Detector module is used to make a reduced state pre-decision on the equalized signal to obtain TD_Seq; The Slicer module is used to make a reduced transition path pre-decision on the equalized signal to obtain S_Seq; The Postfilter module is used to input the equalized signal into the Postfilter to filter out high-frequency noise to obtain PF_Seq; The RST-MLSE module is used to construct the grid graph of the RST-MLSE according to TD_Seq, S_Seq and PF_Seq; The backtracking module is used to backtrack through the grid graph to obtain the estimated maximum likelihood probability of the transmitted signal sequence.

Images