Inputs type of Neural_Receiver for SIMO OFDM System

[nnk14102002]

Hi all,
I have a question about Neural_Receiver for SIMO OFDM System.
Currently, the model input is receiving a tensor with a fixed size full of resource grid.
So when the input of the decoder block changes depending on the configuration at the transmitter (ex: 20RB, 30RB or 50RB ...)(in a slot 14 symbols). What do you think about a solution to divide a large block into many small parts of 1 or 4RB x 14 symbols to fix the input size for the model while still adapt to the change in the size of the input data block.(That means we only train the model with input size which can be only 4RBx14symbols)
Does this solution ignore some correlation information (e.g. about the transmission channel) between RBs in the same large allocated data block like when using full resource grid? This makes it difficult for the model to achieve convergence?

[SebastianCa]

Hi @nnk14102002,
Your solution looks good and aligns with our approach in the Multi-user Neural Receiver.
Training can be performed with a relatively small number of PRBs, while inference can be done using a different configuration.

[nnk14102002]

Hi @SebastianCa ,
Your approach in Multi-user Neural Receiver is a great idea and coincidentally this is also something I am thinking about and want to experiment with. I will read and experiment more on this idea.

Thanks for your help