Over-the-Air Computation with Quantized CSI and Discrete Power Control Levels

Algorithm 1

AirComp-DRL Training for MSE Minimization.

 Require:

 Ensure:

  Initialize , and

  Initialize a Replay Memory

  Initialize action-value function with random weights

  Load -level Channel Quantizer

  fordo

   Draw Channel Coefs

   Assign Power Levels randomly

   ,      Initial State Quantization

   whiledo

    Select a random number

    ifthen               Exploration

     Select a random action

    else                 Exploitation

    end if

    Take action , Observe reward and state

    Store transition in   Experience Replay

    Select random minibatch of transitions from

    Set

    Perform Gradient Descent on

   end while

            -greedy decaying

 end for