Blind Federated Learning at the Wireless Edge With Low-Resolution ADC and DAC

We study collaborative machine learning systems where a massive dataset is distributed across independent workers which compute their local gradient estimates based on own datasets. Workers send through multipath fading multiple access channel with orthogonal frequency division multiplexing to mitigate the selectivity of channel. assume that there no state information (CSI) at workers, and parameter server (PS) employs antennas align received signals. To reduce power consumption hardware costs, we employ complex-valued low-resolution digital-to-analog converters (DACs) analog-to-digital (ADCs), transmitter receiver sides, respectively, effects practical low-cost DACs ADCs performance. Our theoretical analysis shows impairments caused by ADCs, including those one-bit do not prevent convergence federated algorithms, vanish when sufficient number are used PS. also validate our results via simulations, demonstrate using low-resolution, even one-bit, causes only slight decrease in accuracy.