부산대학교 도서관

Designing unimodular waveforms with good auto- and cross-correlation is one of the main problems for multiple-input multiple-output (MIMO) radar. Although long MIMO radar sequences improve long-range detection performance, it is difficult to design such long sequences owing to limitations of code theory and the cost of exhaustive searches. An autoencoder is a type of neural network that has had great success in designing long code sequences for wireless communication. This paper proposes a framework for designing long MIMO sequences with good auto- and cross-correlation properties using an autoencoder. Numerical results show that sequences designed using the proposed framework demonstrate better correlation properties than existing algorithms.