Estimation of transmitter I/Q imbalance using convolutional neural networks

Transmitter induced I/Q imbalance occurs when the modulator's in-phase and quadrature components are not orthogonal. This causes the real and imaginary components of the complex signal to interfere with each other. In addition to potentially degrading performance of the transmitter, I/Q imbalance can also be used as an identifying feature when performing Specific Emitter Identification (SEI) techniques. This work investigates Convolutional Neural Networks (CNNs) as a means to estimate transmitter I/Q imbalance using only raw I/Q data as input instead of predefined features, with no need for synchronization, demodulation, or test signals, for I/Q imbalance correction, SEI, or other applications. The selection of the CNN architecture and network design and training are discussed. Performance analysis shows the network's ability to estimate both gain and phase offset, with performance improving as the number of inputs to the network increases and/or as SNR increases.