Certain aspects of the present disclosure provide techniques for sounding and precoding uplink transmissions using one or more machine learning models. An example method generally includes generating a sounding reference signal (SRS) using an SRS deep neural network (DNN); transmitting, to a network entity, the generated SRS in one or more resource elements (REs); receiving, from the network entity, information identifying a precoding matrix to use for uplink transmissions on a shared channel; precoding uplink transmissions based on the identified precoding matrix; and transmitting, to the network entity, the precoded uplink transmissions on the shared channel.