DEEP LEARNING BASED METHOD AND DEVICE FOR NOISE SUPPRESSION AND DISTORTION CORRECTION OF ANALOG-TO-DIGITAL CONVERTERS

• US 20190319633A1
• Filed: 09/14/2018
• Published: 10/17/2019
• Est. Priority Date: 04/14/2018
• Status: Active Grant

• Current Assignee: Shanghai JIAO TONG University
• Original Assignee: Shanghai JIAO TONG University
• Inventors: ZOU, Weiwen, XU, Shaofu, CHEN, Jianping
• Granted Patent: US 10,812,095 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 18/214 Generating training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06F 18/2414 Smoothing the distance, e.g. radial basis function networks [RBFN]
• CPC Class Codes: G06F 2218/04 Denoising
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/048 Activation functions
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: H03M 1/08 of noise H03M1/0617 takes precedence
• CPC Class Codes: H03M 1/1033 over the full range of the converter, e.g. for correcting differential non-linearity
• CPC Class Codes: H03M 1/1038 by storing corrected or correction values in one or more digital look-up tables H03M1/1057 takes precedence
• CPC Class Codes: H03M 1/12 Analogue/digital converters H03M1/001 – H03M1/10 take precedence
• CPC Class Codes: H03M 1/1215 using time-division multiplexing

Abstract

A method for noise suppression and distortion correction of analog-to-digital converters based on deep learning that realizes effect of correcting noise and distortion of analog to digital converters. The method is applied to electronic ADCs or photonic ADCs. It utilizes the learning ability of the deep network to perform system response learning on ADCs which need noise suppression and distortion correction, establishes a computational model in the deep network that can suppress the reconstruction of noises and distorted signals, performs noise suppression and distortion correction on the signals obtained by ADCs, and thereby improves performance of the learned ADCs. The present invention has a very important role in improving the performance of the microwave photon system with high sampling precision of microwave photon radar and optical communication system.

Claims

1. A device for noise suppression and distortion correction of analog-to-digital converters (ADC) based on deep learning, comprising

an ADC having an input port and an output port, and

a deep learning information processing module, the deep learning information processing module comprising

a microwave signal source having a first output port and a second output port,

a digital signal processor having an input port and a first output port, and

a deep network having a first input port and a second input port,

wherein the first output port of the microwave signal source is connected to the input port of the ADC;

the second output port of the microwave signal source is connected to the input port of the digital signal processor;

the output port of the ADC is connected to the first input port of the deep network; and

the first output port of the digital signal processor is connected to the second input port of the deep network.

2. The device for noise suppression and distortion correction of analog-to-digital converters based on deep learning of claim 1, wherein the ADC is an electronic ADC or a photonic ADC, and the electronic ADC is a data acquisition board or an oscilloscope.

3. The device for noise suppression and distortion correction of analog-to-digital converters based on deep learning of claim 1, wherein the microwave signal source is a phase locked frequency-multiplied source, a voltage controlled oscillator, or an electro-optical oscillation source.

4. The device for noise suppression and distortion correction of analog-to-digital converters based on deep learning of claim 1, wherein the digital signal processor is an field programming gate array (FPGA) or a digital signal processor (DSP).

5. The device for noise suppression and distortion correction of analog-to-digital converters based on deep learning of claim 1, wherein the deep network is a convolution neural network or a recurrent neural network, which can be constructed in a central processing unit (CPU), a graphic processing unit (GPU), a tensor processing unit (TPU), or a photonic neural network.

6. A method for noise suppression and distortion correction of analog-to-digital converters employing the device of claim 1, comprising:

(1) training Phase

inputting a series of standard sinusoidal signals with known frequency and amplitude provided by the microwave signal source respectively into the ADC and the digital signal processor,

converting the standard sinusoidal signals into digital signals with noise and distortion by the ADC, and inputting the digital signals into the deep network,

generating standard digital signals according to the frequency and amplitude information of the standard sinusoidal signals by the digital signal processor, and inputting the standard digital signals into the deep network,

training parameters of the deep network by using the digital signals with noise and distortion and the standard digital signals and applying optimization algorithm for tens of thousands of iterative calculation training, and forming parameters of noise suppression and distortion correction in the deep network for minimizing difference between digital signals with noise and distortion and the standard digital signals, whereby digital signals without noise and distortion are output after being processed based on the parameters of noise suppression and distortion correction in the deep network; and

(2) in an application phase,

inputting analog signals to be sampled through the input port of the ADC,

inputting the converted digital signals with noise and distortion by the ADC into the deep network, wherein the digital signals with noise and distortion is processed based on the parameters of noise suppression and distortion correction, and

outputting digital signals after being process of noise suppression and distortion correction.