Mismatch Error Calibration in Time Interleaved Acquisition System Based on Multi-Head Attention Mechanism

High-speed acquisition systems have wide applications in various scenarios and are important scientific instruments. Due to the rapid growth of the bandwidth of the signals to be acquired, the bandwidth requirements for high-speed acquisition systems also increase. Time-interleaved and digital bandwidth-interleaved architectures make the system more complex, which makes the analysis and correction of instrument errors more difficult. In this paper, we propose a neural network-based digital error correction method and design a parallel acquisition system based on it. The specific steps of the correction method are: using an arbitrary wave generator to synthesize a standard signal, forming a training dataset with the standard signal and the acquisition result containing errors, training the neural network, and deploying it in the digital signal processing module. The experiment shows that the correction result can improve the spurious-free dynamic range of the signal significantly.