Estimating delay spread and doppler spread

An apparatus, comprising:

• at least one processor;
• and
• at least one memory including computer program code, the at least one memory and computer program code configured to, with the at least one processor, cause the apparatus at least to perform:
• initializing trainable parameters for at least one model modeling at least one of a delay spread estimator or a Doppler spread estimator and outputting spread estimations;
• sampling a batch of examples from a dataset comprising data representing data received, corresponding delay spread or corresponding Doppler spread, wherein sampled data in the batch of examples forms a batch of training data and the corresponding spreads form a batch of validation data;
• inputting the batch of the training data to the at least one model;
• updating the trainable parameters by applying a stochastic gradient descent on a loss function, which uses spread estimations outputted by the at least one model and spreads in the batch of validation data;
• training the at least one model by repeating the sampling, inputting and updating until stop criteria are fulfilled;
• and
• storing, when the stop criteria have been fulfilled, the at least one model for spread estimations or for Doppler spread estimations to be used in apparatuses receiving data over wireless channels.

A method, comprising:

• receiving data over a wireless network;
• determining delay spread estimations and Doppler spread estimations by inputting data representing received data to at least one trained model which outputs spread estimations;
• and
• reconstructing the received data based on the data representing received data, which is input into one or more detectors, and the outputs of the delay spread estimations and the Doppler spread estimations.