I/Q Imbalance Compensation Algorithm based on Neural Networks

This paper proposed an I/Q imbalance compensation algorithm based on neural networks, suitable for low-IF receivers. First, the low-IF receiver architecture and the phenomena of I/Q imbalance (also referred as image interference) are described. The standard solution - using a complex LMS adaptive filter, which separates the desired, and image signals - is limited in that the recovered signal remains affected by the I/Q imbalance; the filter proposed here corrects this drawback. The functionality, convergence and stability of the neural network based filter are demonstrated through extensive computer simulations. A sizing example is also given - deduction of the number of sample necessary in order to achieve a -60 dB image rejection - along with the time domain behaviour of the resulting neural network.