In modern wireless communication devices cost- and area-effective signal processing architectures are essential. Flexible and reconfigurable front-end solutions are necessary to achieve high spectral efficiency. Direct conversion transceivers are suitable but have to deal with I/Q mismatch, which, for large bandwidths becomes frequency selective. An example for such impairments are differences between the analog I- and Q-branch low-pass filters due to production tolerances. In order to keep costs low the specifications for the analog components can be reduced, if the compensation of the resulting impairments is done by digital signal processing. Without the knowledge of any transmit data, blind I/Q mismatch compensation can be based on a second-order statistical property called properness, which is valid for a large class of digitally modulated communication signals and which is destroyed by I/Q mismatch.
In a detailed analysis the impact of other RF-impairments on the properness measure is investigated. Under realistic impairment levels we prove that nonlinear even-order distortions resulting from finite mixer isolation have only negligible influence on properness. We propose a novel DSP-algorithm for blind adaptive I/Q mismatch compensators using only real-valued filters, which rebuilds this properness in two stages. Additionally, we add a simple solution which makes the I/Q mismatch compensator insensitive to DC offset. The proposed algorithm outperforms other state-of-the-art algorithms while its computational complexity is reduced. Results from a 3GPP LTE downlink simulator support the analysis. A stability analysis based on Liapunov functions exhibit coupled conditions for the used step-sizes and those compensator parameter spaces for which the optimal steady-state is asymptotically stable. The higher the compensator filter order, the smaller is the compensator parameter space. With fixed starting point and increasing filter order the compensator thus becomes increasingly unstable. Practical filter mismatch orders however, are usually not higher than five.