Brief research of traditional and AI-based models for IMD2 cancellation (2406.09531v1)

Abstract: Due to the limited isolation of duplexer's stopband transceivers operating in frequency division duplex (FDD) encounter a leakage of the transmitted signal onto the receiving path. Leakage signal with the combination of the second-order nonlinearity of the low noise amplifier (LNA) and receiver down-conversion mixer may lead to second-order intermodulation distortion (IMD2) generation thus greatly reducing the receiver sensitivity. Cancellation of undesirable interferences based on adaptation of traditional models such as memoryless and memory polynomials, spline polynomial based Hammerstein and Wiener-Hammerstein models proved its efficiency in case of well-known nonlinearity nature. On the other hand, currently there is an intensive research in the field of nonlinearity detection by means of neural network (NN) structures. NN-based IMD cancellers are effective in the case of unknown interference content due to their high generalization ability. Therefore, NN approach can provide universal model, which is capable of IMD suppression even in case it is hard to separate intermodulation products generated by LNA, down-conversion mixer or even power amplifier in transmitter path. Nevertheless, such structures suffer from high complexity and can`t be implemented in hardware. Current paper presents low-complexity feed-forward NN-based model, which successfully competes with traditional architectures in terms of computational complexity. The testbench results demonstrate the acceptable performance of provided model, which can be equal to the polynomial nonlinear canceler's performance at a reduced computational cost. Current paper provides performance and required resources comparison of traditional memory polynomial-based scheme and NN-based model for IMD2 cancellation.