A/D Converter Background Calibration Algorithm Based On Neural Network

As a combination of chip technology and digital technology,ADC can convert the continuous analog signals in real life,such as optical radiation signal,thermal radiation signal,electrical signal,pressure signal,through electronic circuit technology,through digital processing to discrete digital signals that can be processed by computer.Due to the continuous progress of semiconductor technology,the performance of digital chips has been steadily improved,while power consumption has gradually declined due to the progress of technology,but traditional ADC can not benefit from the decreasing intrinsic gain of transistors and the increasing noise.Firstly,the structure and output performance index of ADC are introduced,and the working principle of ADC and various types of ADC are briefly described.Combined with the requirements of this subject,several error factors affecting the output performance of ADC are analyzed.At the same time,the development and current research status of neural network are introduced,and the feasibility of this topic is illustrated by combining the advantages of neural network.Secondly,according to the research purpose of this topic,several kinds of key error factors in ADC are calculated in detail by theoretical reasoning,and gain error,capacitance mismatch,time error,misalignment error and noise error parameters are analyzed one by one.On the basis of theoretical analysis,the output models of various error parameters of ADC are modeled by using the platform of MATLAB/Simulink,and their frequency spectrum characteristics are obtained.Thenthe output models of various error parameters are integrated to obtain the final equivalent model of ADC output.Then,based on the previous research of neural network,the overall structure of the correcting neural network is designed,and the structure and training methods of the neural network are selected.The middle layer structure of the corrected neural network is 16-32-16.The input layer and the output layer correspond to ADCs with different effective digits respectively.The activation function is tanh function,the batch number is set to 20,the initial learning rate of the neural network is set to 0.0025,and the comprehensive termination condition of setting iteration times and error values is used.According to the objective of this project,the main error factors of ADC can be corrected by the correcting neural network designed in this project.Finally,the whole correction algorithm is simulated to verify the effectiveness of the whole correction algorithm.In the learning stage,the output error function is used to observe whether the error function can continue to decline and stabilize in the prescribed iteration times,and to verify the normal convergence of the correction neural network.In the correction stage,the convergent network is used to correct the ADC,and the effectiveness of the algorithm is verified by the correction results.The experimental results show that when the input signal frequency is 12 MHZ and the sampling frequency is 100 MHZ,the output characteristic SNDR of ADC is increased from 53.16 dB to 73.11 dB and ENOB is increased from 8.58 Bi to 11.93 bit through the correction algorithm based on neural network.The algorithm can effectively correct the harmonic distortion error of ADC and improve the performance of pipeline ADC.