부산대학교 도서관

Arbitrary waveform generator (AWG) can generate various excitation signals flexibly and is widely used in the automatic testing system (ATS). With the continuous evolution of electronic science and technology, higher demand for the AWG's sampling rate and output bandwidth has been put forward. Frequency-interleaved Digital-to-analog converter (FI-DAC) can improve those parameters quite effectively. However, the phase deviation between sub-band paths in the FI-DAC will cause a severe error in the overlapping band when the sub-band signals are combined. Therefore, we set up the error model for the FI-DAC overlapping band, analyzed the impact of the phase deviation of the output signal, and proposed a phase compensation method based on all-pass filter. The all-pass filter coefficient solution for phase compensation is a non-linear least square (NLS) problem and is usually solved using meta-heuristics. Yet the traditional genetic algorithm (GA) has a slow convergence speed, and the particle swarm optimization (PSO) tends to fall into local optimal when solving for high-order filter coefficient. Hence, we analyzed the parametric characteristics of all-pass filter and proposed a modified GA (MGA) to solve filter coefficients, compensate for the phase deviation between the sub-bands, and guarantee the quality of the final synthesized signals. The experiment result shows that, under the same number of iterations, the root mean square (RMS) error of the traditional GA is 0.1736 rad, PSO is 0.7725 rad, while the error of our MGA is only 0.0387rad, which is significantly better than the conventional method.