ADAPTIVE DEMODULATION METHOD AND APPARATUS USING AN ARTIFICIAL NEURAL NETWORK TO IMPROVE DATA RECOVERY IN HIGH SPEED CHANNELS
A neural network demodulator is used within a receiver to provide Inter Symbol Interference (ISI) channel equalization and to correct for I/Q/phase imbalance. The neural network is trained with a single integrated training step to simultaneously handle the channel impairments of ISI equalization and I/Q phase imbalance as opposed to prior art methods of separately addressing each channel impairment in sequence.
1 . A method for use in a communication system, the communication system including a transmitter and a receiver, to provide ISI channel equalization and to correct I/Q phase imbalance, comprising,
providing a demodulator as part of the receiver, the demodulator including a trainable neural network, and
training the neural network in a single integrated training step to simultaneously enable the demodulator to provide ISI channel equalization and to correct I/Q phase imbalance.
2 . A method in accordance with claim 1 wherein the neural network includes two layers, the
method further comprising the step of implementing a transfer function between the two layers of the neural network to ensure convergence of the neural network training step.
3 . A method in accordance with claim 2 wherein transmitter data symbols and RF carrier
signals are sent over the communication channel from the transmitter to the receiver with the I/Q imbalance resulting from carrier phase misalignment between the
transmitter and receiver, said neural network training step utilizing information sent over the communication channel to determine coefficients needed by the neural network to correct said carrier phase misalignment.
4 . A method in accordance with claim 3 wherein transmitter data symbols and RF carrier
signals are sent over the communication channel from the transmitter to the receiver with ISI resulting by the change of bandwidth of the carrier signal frequency
causing interference between adjacent transmitter data symbols, said neural network training step utilizing information sent over the communication channel to
determine coefficients needed by the neural network to prevent the interference between transmitter data symbols.
5 . A method in accordance with claim 4 wherein the neural network training process
determines coefficients W1, b1 W2 and b2 such that the mean squared error between a transmitted symbol and a symbol received at the output of the demodulator is minimized.
6 . A communications system, comprising,
a transmitter for sending modulated data symbols and carrier signals over the communications channel,
a receiver including a demodulator, and a trainable neural network included as part of the demodulator, which when trained in a single integrated training step allows the demodulator to simultaneously provide ISI channel equalization and correct I/Q phase imbalance.
7 . A communications system in accordance with claim 6 , wherein the neural network
includes two layers, one hidden layer and one output layer, the hidden layer applying a first set coefficients to data symbols from the receiver to create a first set of
fixed point words, and processing the first set of fixed point words with a transfer function.
8 . A communications system in accordance with claim 7 , wherein the first set of fixed
point words are sent to the output layer after processing by the transfer function and applied to a second set of coefficients.
9 . A communications system in accordance with claim 8 , wherein the neural network is
trained based on information sent over the communication channel to generate the first and second set of coefficients.
10 . A communications system in accordance with claim 9 , wherein the first and second set
of coefficients ensure that the mean squared error between a transmitted data symbol and a symbol at the output of the demodulator is minimized.