Denoising The Wireless Channel Corrupted Images Using Machine Learning

Digital communication depends on channel coding for the integrity and correct reception of the data. The traditional communication techniques ignore the context and content associated with the received data while mitigating the channel effects. Image transmission in digital communication have many major constraints. Image quality degrades over wireless channel due to limited characteristic of transmitted data. To mitigate the effects of the noisy channel on the images different image denoising techniques are used such as DCT, Convolutional Denoising Autoencoders, DWT, BM3D, FFT, NLM and Deep Learning. Main objective of this paper is to study the application of machine learning in digital communication to correct errors and remove the effects of channel degradation. This will help us improve the energy consumption, resource utilization, and the Bit Error Rate. It will enable us to communicate with low bandwidth and using minimum resources. Machine learning for denoising the image has attracted substantial attentions because of its high denoising performance. We have constructed a Convolutional Neural Network that will denoise the image which is corrupted by the noisy channel. The existing traditional models only denoise the image for a specific noise type (Gaussian noise) and certain noise level $\sigma =$ 25, but our trained network denoise the image for unknown noise level. By using the methods of batch normalization and residual learning the denoising performance is increased. Experiments prove that our trained model displays high efficiency in image denoising. The proposed network is trained in MATLAB and Python with the help of GPU computing which accelerates neural network’s performance. The application areas of this paper mainly are in the domains related to digital communications. This include wireless communications, optical communications and line communications. However, as the wireless communications (especially underwater wireless) is extremely error prone, so this domain will be highly benefited. The applications can include military communications, Underwater Communication and Noisy Industrial Communication.