-
FBLMS算法通过应用离散傅里叶变换理论的重叠保留法[16-17],对均衡器的输入数据分块处理,将线性卷积转化为循环卷积,利用循环卷积定理,将时域循环卷积转化为频域乘积后,用LMS算法修正频域权系数,从而达到自适应频域均衡的目的。少模光纤通信系统中的多个传输模式、相干接收机构成MIMO系统。在奈奎斯特采样频率下,每个正交相移键控(quadrature phase shift keyin, QPSK)信号被采样两次,所以需要两组均衡器均衡奇偶次序数据。对于具有D个传输模式的少模光纤通信系统,必须保证任意两个模式间均有一个子均衡器,才能从对应模式中提取出相应的耦合分量,此时共需要2D2个子均衡器。
MIMO-FBLMS算法流程如下:对于N阶FBLMS自适应均衡器,模式i的输入信号经2倍符号速率采样后,得到奇次序和偶次序数据,并分为大小为N的数据块。运用重叠保留法后经2N点傅里叶变换得到频域信号。
$ {\mathit{\boldsymbol{X}}_{i, {\rm{o}}}}(k) = {\rm{FFT}}({\mathit{\boldsymbol{x}}_{i, {\rm{o}}}}(k)){\rm{ }} $
(1) $ {\mathit{\boldsymbol{X}}_{i, {\rm{e}}}}(k) = {\rm{FFT}}({\mathit{\boldsymbol{x}}_{i, {\rm{e}}}}(k)) $
(2) 式中, k表示第k次迭代过程,下标o代表奇次序,e代表偶次序。xi表示输入的时域信号,Xi表示与之对应的频域信号,FFT表示快速傅里叶运算(fast Fourier transform)。对于每个模式对ij,都存在两组均衡器Wij, o, Wij, e分别与奇偶次序数据对应,实现该模式对的解复用。
$ \begin{array}{l} {\mathit{\boldsymbol{Y}}_i}\left( k \right) = \sum\limits_{j = 1}^D {} {W_{ij, {\rm{o}}}}^{\rm{T}}(k){\mathit{\boldsymbol{X}}_j}\left( k \right) + \\ \sum\limits_{j = 1}^D {} {\mathit{\boldsymbol{W}}_{ij, {\rm{e}}}}^{\rm{T}}(k){\mathit{\boldsymbol{X}}_j}(k) \end{array} $
(3) $ {\mathit{\boldsymbol{y}}_i}(k) = {\rm{IFFT}}({\mathit{\boldsymbol{Y}}_i}(k)) $
(4) 式中, Wij对应ij模式对的均衡器权系数,上标T表示转置运算,IFFT是逆快速傅里叶运算, Yi是解复用后模式i的频域信号,经过逆傅里叶变换得到其时域信号yi。
$ {\mathit{\boldsymbol{e}}_i}\left( k \right) = {{\mathit{\boldsymbol{\bar y}}}_i}\left( k \right) - {\mathit{\boldsymbol{y}}_i}(k) $
(5) $ {\mathit{\boldsymbol{E}}_i}(k) = {\rm{FFT}}{\left[ {0\;\;\;\; \ldots \;\;\;\;0\;\;\;\;{\mathit{\boldsymbol{e}}_i}\left( k \right)} \right]^{\rm{T}}} $
(6) $ {\nabla _{ij}}\left( k \right) = {\rm{IFFT}}({\mathit{\boldsymbol{E}}_i}\left( k \right)*{\mathit{\boldsymbol{X}}_j}^*(k)){\rm{ }} $
(7) $ \begin{array}{l} {\mathit{\boldsymbol{W}}_{ij}}(k + 1) = {\mathit{\boldsymbol{W}}_{ij}}\left( k \right) + \\ \mu {\rm{FFT}}{\left[ {{\nabla _{ij}}\left( k \right)\;\;\;\;0\;\;\;\;{\rm{ }} \ldots \;\;\;\;{\rm{ }}0} \right]^{\rm{T}}} \end{array} $
(8) 式中, ${\mathit{\boldsymbol{\bar y}}_i}\left( k \right) $是理想信号,训练阶段时为已知的训练序列,自适应阶段时是均衡后信号的判决结果。通过与解复用后的时域信号yi相减,得到时域信号误差ei,通过傅里叶运算后得到频域信号误差Ei,进而修正均衡器权系数。*代表复共轭,$ {\nabla _{ij}}\left( k \right)$是时域权系数增量,μ是迭代步长,无量纲,用来调整算法的迭代速度。
(6) 式~(8)式是在重叠保留法下,通过信号误差修正均衡器权系数的计算式。随着算法迭代次数增加,信号误差越小,均衡器权系数逐渐接近最优值,即算法收敛。
均衡算法收敛与否,可通过均衡后信号的Q2因子和均方误差MSE判断。Q2因子是信号功率与均方误差之比,Q2因子越大,均衡算法效果越好。当算法稳态收敛时,均方误差和Q2因子基本不变。
当D=2,即用双模光纤传输时,FBLMS均衡算法的系统流程图如图 1所示。
对于D×D MIMO均衡器,采用基2分解的傅里叶运算时,时域LMS算法平均计算复杂度为:
$ {C_{{\rm{TDE}}}} = 3D\Delta \tau LR $
(9) 频域FBLMS平均计算复杂度为:
$ {C_{{\rm{FDE}}}} = (4 + 4D){\rm{lo}}{{\rm{g}}_2}(2\Delta \tau LR) + 8D $
(10) 式中, Δτ为模式间最大模式时延,L为系统传输距离,R为信号传输速率,平均计算复杂度CTDE和CFDE用单个模式单个符号所需的乘法次数表示,无量纲。
-
FBLMS算法存在收敛速度和收敛效果无法同时满足的矛盾,使用变步长方法可以克服这个问题。变步长方法的调整原则是:算法初始收敛阶段或信道参量改变时,均方误差较大,则应使用大步长,加快收敛速度和系统追踪速度;算法接近收敛后,权系数接近最佳值,此时均方误差很小,步长也应较小,从而得到小稳态失调:
$ \mu = {\mu _{{\rm{max}}}}\frac{1}{{1 + {{\rm{e}}^{ - \alpha \cdot(\varepsilon - \beta )}}}} $
(11) (11) 式是构造的变步长函数,通过均衡后信号的均方误差来控制步长因子。式中, ε表示均方误差;μmax是常数,控制步长的最大值; α是函数系数,控制曲线形状,α越大,在临界区域步长变化越明显,临界区域也越小; β是平移量,确定临界区域中心位置,临界区域将整个均方误差分为3个部分,大步长区域,小步长区域和临界区域。均方误差大于临界区域时μ较大,算法收敛速度快。在临界区域内,随着均方误差减小,μ随之减小。当均方误差低于临界区域限值时μ很小,保证算法在稳态时有较小的稳态失调。此时μ有一个最小迭代步长,保证算法在稳态时仍有一定追踪能力。每次迭代过程都通过数据块的均方误差来修正权系数,消除了用单个信号误差修正时的偶然性和随机性,确保算法稳定收敛。
少模光纤通信系统中的自适应频域均衡算法
Adaptive frequency-domain equalization for few-mode fiber transmission systems
-
摘要: 少模光纤模式复用存在模式耦合和差分模式时延,必须通过自适应均衡算法补偿。为了降低长距离少模光纤通信系统中自适应均衡算法的复杂度,采用基于变步长-频域块最小均方算法的多输入多输出均衡器对2×2模分复用系统解复用。利用频域块最小均方自适应算法修正均衡器权系数,并通过变步长函数调整步长因子,兼顾算法收敛速度和收敛性能。算法可通过快速傅里叶变换降低计算复杂度。在112Gbit/s的1000km少模光纤高速通信仿真系统中,保证相同收敛速度情况下,提高信号Q2因子3.7dB,并在可编程现场门阵列上验证了100km少模光纤通信系统时的算法性能。结果表明,该算法能够实现模分复用系统的信号解复用,达到快速收敛、低稳态失调的目的。
-
关键词:
- 光通信 /
- 变步长-频域块最小均方算法 /
- 频域均衡 /
- 少模光纤 /
- 模式复用
Abstract: Modal crosstalk and differential mode delay in mode-division multiplexing in few-mode fiber must be compensated for with adaptive equalization algorithm. For the purpose of decreasing complexity of adaptive equalization algorithm in long-haul few-mode fiber communication system, a multiple-input multiple-output equalizer with variable step-size frequency-domain-block least mean square(VS-FBLMS) algorithm was proposed to uncouple a 2×2 mode-division multiplexing system. Compromising the conflicting requirements of convergence speed and convergence performance, FBLMS adaptive algorithm was employed to update equalizer taps and modify step-size factor with variable step-size function. Furthermore, the algorithm decreased the computational complexity by using fast Fourier transform. In a 112Gbit/s high-speed communication simulation system over 1000km in few-mode fiber, signal Q2 factor increased 3.7dB with same convergence speed. The performance of the algorithm was tested on field programmable gate array with a 100km few mode fiber communication system. The study demonstrates that the algorithm successfully uncouples the signals of mode-division multiplexing system and realizes fast convergence and low disorder. -
[1] RYF R, RANDEL S, GNAUCK A H, et al. Mode-division multiplexing over 96km of few-mode fiber using coherent 6×6 MIMO processing[J]. Journal of Lightwave Technology, 2012, 30(4):521-531. doi: 10.1109/JLT.2011.2174336 [2] BAI N. Mode-division multiplexed transmission in few-mode fibers[D]. Orlando, USA: University of Central Florida Orlando, 2013: 9-20. [3] YANG J, LIU M, ZHU M, et al. Three-core photonic crystal fiber with zero intermodal dispersion[J]. Laser Technology, 2015, 39(4):528-532(in Chinese). [4] QIAN Y, LIU M, YANG J, et al. Analysis of coupling characteristics of square five-core photonic crystal fibers[J]. Laser Technology, 2014, 38(4):455-458(in Chinese). [5] HU G J, WANG Y P, RUI L, et al. Mode demultiplexing based on CMA under the condition of differential group delay[J]. Journal of Jilin University Engineering and Technology Edition, 2015, 45(3):961-965(in Chinese). [6] GRUNER-NIELSEN L, SUN Y, NICHOLSON J W, et al. few-mode transmission fiber with low DGD, low mode coupling and low loss[J]. Journal of Lightwave Technology, 2012, 30(23):3693-3698. doi: 10.1109/JLT.2012.2227243 [7] FERREIRA F, FONSECA D, LOBATO A, et al. Reach improvement of mode division multiplexed systems using fiber splices[J]. IEEE Photonics Technology Letters, 2013, 25(12):1091-1094. doi: 10.1109/LPT.2013.2256120 [8] CAO X. Optimization of dispersion compensation in optical fiber communication systems[J]. Laser Technology, 2014, 38(1):101-104(in Chinese). [9] FAN Zh, WEN G Q, ZHOU H, et al. Research on dispersion compensation for OFDM signal fiber transmission[J]. Laser Technology, 2011, 35(1):112-116(in Chinese). [10] UDRN R G H, OKONKWO C M, SLEIFFER V, et al. Adaptive step size MIMO equalization for few-mode fiber transmission systems[J]. European Conference & Exhibition on Optical Communication, 2013, 22(1):1-3. [11] RANDEL S, WINZER P. DSP for mode division multiplexing[C]//Optoelectronics and Communications Conference Held Jointly with 2013 International Conference on Photonics in Switching (OECC/PS), 201318th.New York, USA: IEEE, 2013: 1-2. [12] YAMAN F, BAI N, ZHU B Y, et al. Long distance transmission in few-mode fibers[J]. Optics Express, 2010, 18(12):13250-13257. doi: 10.1364/OE.18.013250 [13] BAI N, LI G F. Adaptive frequency-domain equalization for mode-division multiplexed transmission[J]. IEEE Photonics Technology Letters, 2012, 24(21):1918-1921. doi: 10.1109/LPT.2012.2218802 [14] RANDEL S, WINZER P J, MONTOLIU M, et al. Complexity analysis of adaptive frequency-domain equalization for MIMO-SDM transmission[C]//Optical Communication (ECOC 2013), 39th European Conference and Exhibition on IEEE.New York, USA: IEEE, 2013: 1-3. [15] VUONG J, RAMANTANIS P, SECK A, et al. Understanding discrete linear mode coupling in few-mode fiber transmission systems[C]//Optical Communication (ECOC), 201137th European Conference and Exhibition on IEEE.New York, USA: IEEE, 2011: 1-3. [16] SHI J. Study on mode division multiplexing technology in few-mode fiber[D]. Changchun: Jilin University, 2013: 17-47(in Chinese). [17] FARUK M S, KIKUCHI K. Adaptive frequency-domain equalization in digital coherent optical receivers[J]. Optics Express, 2011, 19(13):12789-12798. doi: 10.1364/OE.19.012789 [18] BAI N, IP E, LI M J, et al. Long-distance mode-division multiplexed transmission using normalized adaptive frequency domain equalization[C]//IEEE Photonics Society Summer Topical Meeting Series.New York, USA: IEEE, 2013: 135-136.