The federated learning-based symbol detection technique has attracted much attention recently due to its application of processing data at the network edge (ie, mobile user equipment) in a distributed manner. In this regard, the federated receiver is presently being promoted as a replacement for and improvement on the conventional model-based receiver. This paper proposes an average pooling with compensation layer and an average pooling with dense layers-added federated receiver for symbol detection in a downlink fading channel. In average pooling with a compensation scheme, the average of the input data is extracted and then ripples of the data are minimized by multiplying the weight coefficient in the compensation layer. In average pooling with dense scheme, two pooling schemes, such as average pooling and max-pooling, are incorporated into dense layers with various activation functions. Simulation results are provided to demonstrate that the proposed compensation layer-added federated receiver outperforms the conventional federated receiver.

No abstract