Orthogonal frequency-division multiplexing equalization using deep neural network

An apparatus comprising:

• a memory storage;
• and
• a processing unit coupled to the memory storage, wherein the processing unit is operative to:
• receive, at an Orthogonal Frequency-Division Multiplexing (OFDM) receiver, a signal in a packet structure from an OFDM transmitter, the signal having distortion;
• extract training constellation points, pilot constellation points, and data constellation points from the signal based on the packet structure, wherein each data constellation point corresponds to a data subcarrier within a data symbol of the signal;
• provide the training constellation points and the pilot constellation points as input for the data symbol to a Deep Neural Network (DNN);
• receive as output from the DNN a coefficient for each data subcarrier within the data symbol that reverses the distortion;
• and
• apply the coefficient for each data subcarrier to the corresponding data constellation point to determine a per subcarrier constellation point prediction.

A method comprising:

• receiving at an Orthogonal Frequency-Division Multiple Access (OFDMA) receiver:
• a first signal in a packet structure transmitted by a first OFDMA transmitter over a first subchannel of a communication channel, the first signal having distortion;
• and
• a second signal in the packet structure transmitted by a second OFDMA transmitter over a second subchannel of the communication channel, the second signal having distortion;
• for each of the first signal and the second signal, extracting training constellation points, pilot constellation points, and data constellation points from the respective signal based on the packet structure, wherein each data constellation point corresponds to a data subcarrier within a data symbol of the respective signal;
• and
• processing at least a portion of the first signal and the second signal separately to account for a difference in the distortion of the first signal and the second signal, wherein the processing for each of the first signal and the second signal includes:
• providing the training constellation points and the pilot constellation points extracted from the respective signal as input for the data symbol to a Deep Neural Network (DNN);
• receiving as output from the DNN a coefficient for each data subcarrier within the data symbol that reverses the distortion of the respective signal;
• and
• applying the coefficient for each data subcarrier to the corresponding data constellation point to determine a per subcarrier constellation point prediction, wherein the per subcarrier constellation point prediction predicts an original data constellation point corresponding to each data subcarrier at the respective OFDMA transmitter prior to transmission of the respective signal.