Research On Demultiplexing Technology For Mode Division Multiplexing System

With the global data traffic increasing exponentially, the capacity of the single-mode fiber communication system is approaching its nonlinear Shannon limit. As a new and promising capacity expanding method, mode division multiplexing(MDM) technology can multiply the capacity of optical fiber communication system by exploring the modes as a new degree of freedom for transmission. However, in a practical MDM system based on few-mode fiber, mode coupling and differential mode group delay(DMGD) will deteriorate the transmission performance severely and thus limit the transmission distance. With the interplay between mode coupling and DMGD, the transmission process of the system is far more complex than the single-mode fiber system resulting that each output signal of the system is a superposition of multiple linear convolutions between input signals and channel impulse response. Therefore, it is urgently required to develop an effective demultiplexing method for MDM system to compensate the impairments caused by mode coupling and DMGD and demultiplex the output signals.In this dissertation, mode demultiplexing methods for MDM system are investigated by theoretical derivation and simulated verification. We analyze the transmission characteristics of the modes, the generation mechanism of different impairments and their effect mechanism on the modes, and the transmission model of the MDM system is deduced according to the matrix propagation model. Based on the analysis above, the blind signal processing such as multimodulus blind equalization algorithm(MMA) and multichannel blind deconvolution(MBD) are induced to conduct the impairments compensation and implement the mode signals separation and eventually achieve mode demultiplexing.The main contributions and the innovative points of this dissertation are summarized as follows.1. Considering that the frequently-used blind equalization algorithms have a high steady-state error for high-order modulation format, a demultiplexing method based on time-domain multimodulus blind equalization algorithm(TD-MMA) is proposed. The multiple-input multiple-output MMA suitable for MDM system is constructed through an extension of the single-input single-output MMA. And the simulation system was built to verify the performance of TD-MMA for low-order and high-order modulation formats respectively. The results show that TD-MMA can achieve demultiplexing without the carrier phase recovery algorithm for different modulation formats. In addition, considering that the proposed TD-MMA shows a high computational complexity in long-haul MDM transmission, a demultiplexing method based on frequency-domain multimodulus algorithm(FD-MMA) is proposed. First of all, butterfly-structured equalizers including even and odd sub-equalizers are built to satisfy the twofold-oversampled input signals and the block processing. Then fast Fourier transformation(FFT) is implemented to extend time-domain MMA into frequency-domain. Finally, the gradient update rule of FD-MMA is deduced. The results show that the proposed FD-MMA significantly reduces the computational complexity compared with TD-MMA while maintaining a similar performance, meanwhile, its computational complexity is in the same order of magnitude with the frequency-domain least mean square(FD-LMS) algorithm.2. Considering the special channel transmission characteristic of the MDM system, we propose a mode demultiplexing method based on multichannel blind decinvolution(MBD) which separates the convolutive mixing outputs in the time-domain directly. In the MBD algorithm, the complex convolutive mixing is reformulated as a relatively simple instantaneous mixing by extending the received signals to their time-delayed versions, and then the resulted instantaneous mixtures can be separated by the mature and efficent ICA algorithm. Simulation results show that compared to the frequently-used equalization algorithms, the proposed MBD achieves similar demultiplexing performance in the case of low-order modulation formats and a better performance in the case of higher-order formats. Furthermore, considering the high computational complexity of the proposed time-domain MBD, the frequency-domain multichannel blind deconvolution, namely the frequency-domain independent component analysis(FD-ICA) algorithm, is proposed in this dissertation. Firstly, FD-ICA transforms the convolutive mixture in the time-domain to the instantaneous mixture in the frequency domain through short-time Fourier transform(STFT). After time-frequency transformation, the frequency components of the mode signals in each frequency bin are separated by the complex-valued ICA and then the permutation algorithm is utilized to eliminate the residual permutation ambiguity of the ICA. Finally, the time-domain estimated signals are obtained via inverse STFT. According to the complexity analysis, the computational complexity of FD-ICA is reduced significantly and insensitive to the number of transmission modes. However, the computational complexity of FD-ICA will increase sharply when the total DMGD of the link reach a certain value.3. Considering that the complexity of the demultiplexing algorithm based on FD-ICA shows a sharply increase trend with the increasing of the DMGD, an improved continuous convergence FD-ICA(ICCFD-ICA) algorithm is proposed. We illustrate the source of the permutation ambiguity by theoretical derivation and verify the specific impact of the permutation ambiguity on the demultiplexing performance by the simulation experiments. For avoiding the computational burden of the permutation algorithm, we propose a new initialization method for the separation matrix of each frequency bin according to the characteristic that the separation matrices of adjacent frequency bins are similar. In this method, the converged separation matrix in the lower frequency bin is employed as the initial value of the separation matrix in the adjacent higher frequency bin, therefore, the convergent tendency of the separation matrix of the higher frequency bin can be controlled and the instantaneous separation system of adjacent frequency bins will tend to converge in the continuous direction and the resulting same permutation order. As a result, the step of solving the permutation ambiguity can be avoided. The complexity anslysis results show that the proposed ICCFD-ICA has lower computational complexity than both FD-LMS and FD-ICA. Moreover, the simulation results show that ICCFD-ICA is a fast and efficient demultiplexing algorithm showing a similar performance with FD-LMS and FD-ICA, however, its convergence speed is improved by 54.5% with respect to FD-ICA.In this dissertation, combined with the special channel characteristics of MDM system, we propose some new demultiplexing methods based on MMA and MBD. The proposed algorithms have the advantages in increasing the throughput efficiency of the system, mitigating the phase noise, demultiplexing the signals with high-order formats, as well as reducing the computational complexity, which will provide some reference for the further research on the demultiplexing technology for MDM system.