Non-orthogonal multiple access (NOMA) has been
introduced as a promising scheme to allow for superposition
of signals, such as the transmission of multiple services in the
same resource block (time and frequency). In this paper, we
propose the application of Deep Learning (DL) in an Autoencoder
(AE) framework for simultaneous usage in multi-service NOMA
transmission. While classical NOMA simultaneously incorporates
locally separated user equipments (UEs), we focus on the simul-
taneous transmission of services within a single UE. Our scheme
achieves improved performance compared to classical NOMA
schemes and is capable of performing optimal estimation of the
superimposed transmit signals. The scheme utilizes an equidistant
power allocation scheme. The results show the potential of using
DL to enhance the performance of NOMA systems and improve
their adaptability and flexibility to different scenarios.