Multiple Input Multiple Output (MIMO) and massive MIMO (mMIMO) are key-enabling technologies for 4G and 5G communications systems. mMIMO uses a high number of antennas, where the number of antennas at the base station exceeds in general the number of antennas in the mobiles. For uncorrelated channels, linear equalizers already achieve promising performance in the uplink due to the channel hardening effect. In contrast, we will focus on large symmetrical MIMO systems in 

this paper, where many antennas are employed at the transmitter and the receiver side resulting in a more challenging task for the receiver. Traditionally, receiver algorithms have been derived based on models for the communications system. Recently, machine learning approaches have been proposed where the design is data driven. In order to overcome the drawbacks of model-based and pure data driven approaches, hybrid approaches combining the benefits of both worlds have emerged. In this paper, we present the novel hybrid approach entitled Deep Equalization (DeEQ) based on model knowledge and a neural network like structure. As demonstrated by simulation results, this novel approach achieves very good performance with the advantage of only a very low error floor.