Abstract:Massive multiple-input multiple-output (MIMO) is esteemed as a critical technology in 6G communications, providing large degrees of freedom (DoF) to improve multiplexing gain. This paper introduces characteristic mode analysis (CMA) to derive the achievable DoF. Unlike existing works primarily focusing on the DoF of the wireless channel,the excitation and radiation properties of antennas are also involved in our DoF analysis, which influences the number of independent data streams for communication of a MIMO system. Specifically, we model the excitation and radiation properties of transceiver antennas using CMA to analyze the excitation and radiation properties of antennas. The CMA-based DoF analysis framework is established and the achievable DoF is derived. A characteristic mode optimization problem of antennas is then formulated to maximize the achievable DoF. A case study where the reconfigurable holographic surface (RHS) antennas are deployed at the transceiver is investigated, and a CMA-based genetic algorithm is later proposed to solve the above problem. By changing the characteristic modes electric field and surface current distribution of RHS, the achievable DoF is enhanced. Full-wave simulation verifies the theoretical analysis on the the achievable DoF and shows that, via the reconfiguration of RHS based on the proposed algorithm, the achievable DoF is improved.
Abstract:Domain adaptation aims to enable Large Language Models (LLMs) to generalize domain datasets unseen effectively during the training phase. However, factors such as the size of the model parameters and the scale of training data are general influencers and do not reflect the nuances of domain adaptation performance. This paper investigates the fine-grained factors affecting domain adaptation performance, analyzing the specific impact of `words' in training data on summarization tasks. We propose quantifying dataset learning difficulty as the learning difficulty of generative summarization, which is determined by two indicators: word-based compression rate and abstraction level. Our experiments conclude that, when considering dataset learning difficulty, the cross-domain overlap and the performance gain in summarization tasks exhibit an approximate linear relationship, which is not directly related to the number of words. Based on this finding, predicting a model's performance on unknown domain datasets is possible without undergoing training.