Improved Differential Evolution Algorithms And Its Applications In Communication Signal Processing

With the promotion of computer technology, the intelligent optimization technique becomes mature gradually. It has been developed as an important application subject and increasingly revealed effects in understanding and transforming the world. Proved by engineering applications, the performance of a system can be enhanced significantly with an optimizing procedure. In communication and signal processing area, many problems can boil down to a kind of optimization issue in nature. For instance, communication waveforms shaping and optimizing design, features extraction of modulation signals, and symbol detection in MIMO (multi-input multi-output) communication system, etc.. These problems usually can not be solved ideally by traditional methods. However, we can obtain their approximate optimal solutions through an optimization process. As a new search technique, differential evolution has been profoundly researched and widely used for its advantages of simple and particle uses, fast convergence, and robust. It has made special contributions to the theoretical innovation and technological development of evolutionary computation.However, similar to other population based optimizers, differential evolution has certain drawbacks such as stagnation and premature convergence. Particularly for high-dimensional complex problems, the standard differential evolution can not solve them effectively. To overcome the defects of standard differential evolution algorithm, improving ways have been proposed and further tested over benchmark functions in this dissertation. Finally, two specific applications in communication area have been studied based on the improved differential evolution algorithm. The main contributions of the work are as follows:(1) The dissertation proposes an effective memetic differential evolution algorithm, or DECLS, that utilizes parameter adaptation to enhance the global optimizing performance, and chaotic local search to improve the final solution accuracy by carefully exploiting around the best individual. Moreover, the randomness of chaotic search can compensate for the premature convergence of standard DE to some extent. Simulations proved that the combination of a chaotic local search and a parameter adaptation mechanism is very reasonable. Results show that DECLS is superior to the standard DE and other DE variants. What is more, the DECLS has also shown certain advantages in solving high dimensional problems.(2) To overcome the problem that standard DE can not be directly used in a binary search space, the dissertation proposed an adaptive binary Differential Evolution algorithm, or ABDE, that improves the mutation strategy and adaptively controls the scaling and crossover factors to obtain a better optimization result. Experiments have been carried out by comparing ABDE with two binary DE variants and the most used Genetic Algorithm on a set of 13 selected benchmark functions and the classical 0-1 knapsack problem. Results show that the ABDE performs better than, or at least comparable to, the other algorithms in terms of search ability, convergence speed, and solution accuracy.(3) By analyzing the Very Minimum Chirp Keying (VMCK) modulation signal, the dissertation proposed an optimization scheme that utilizes the sinusoidal basis fitting and differential evolution to modify the spectrum shapes. The theoretical analysis and simulation show that the proposed scheme can remove the harmonic spectral lines successfully and obtain an optimized VMCK waveform with narrower bandwidth, lower spectrum sidebands, and better demodulation performance.(4) By transforming the minimization of bit error rate of a Multi-input Multi-output (MIMO) communication system into an optimization problem, the dissertation proposed a new linear MIMO symbol detector that utilizes the memetic differential evolution to optimize the MBER detector's coefficients. Simulation shows that the proposed detector is superior to MMSE and ZF linear detectors. Further, to save the computation consumption of ML detector under the allowed bit error rate, the dissertation presents a new non-linear MIMO symbol detection method which uses a binary differential evolution optimization to replace the exhaustive search technique.