Simultaneous Echo Cancellation and Upstream Signal Recovery using Deep Learning in Full-duplex DOCSIS Systems (2020)

By Qi Zhou, You-Wei Chen, Shuyi Shen & Gee-Kung Chang, Georgia Institute of Technology; Jeff Finkelstein, Drew Davis & Brian Lee, Cox Communications,

Full-duplex DOCSIS has been encountering a continuous uphill battle to double the channel capacity through resource sharing of uplink and downlink channels. As the downstream and upstream are delivered via the same spectrum at the same time, co-channel interference in the form of internal coupling, micro-reflections or echoes becomes a formidable challenge. To ensure the proper operations of full-duplex DOCSIS, echo cancellation is an urgently needed technique. Typically, it involves analog cancellation to lower the power level of the major echoes below the analog-to-digital converter (ADC) dynamic range, while digital cancellation is followed to remove the residual echoes ensuring a sufficient modulation error ratio (MER).

Many conventional echo cancellers are realized via a subtraction scheme, which implies the receiver is operating linearly. However, this limits the transceiver operation margin and the achievable MER. As the power of the desired and the echo signal increase, the receiver front-end may be driven away from its linear operation range and introduce nonlinear impairments. The crosstalk among the echoes and the desired upstream signals would degrade the echo cancellation performance. Thus, a simple subtraction based cancellation is no longer sufficient.

In this paper, we propose a deep neural network (DNN) based method to simultaneously cancel the echoes and recover the upstream signal. Both the received signal and the known downstream signal will be fed into the DNN processer, and the DNN output is the recovered upstream signal. The DNN is an efficient method to mitigate nonlinearities because of the implemented nonlinear activation function at each hidden layer. After proper initial training, the DNN-based canceller can achieve an excellent upstream signal recovery and outperform the conventional digital cancellation schemes. Moreover, the on-demand dynamic training can be performed implicitly without impacting the regular DOCSIS system operation. This novel approach would dramatically improve the recovered upstream signal quality and increase the capacity of the DOCSIS. This paper is organized as follows. Section 1 reviews the background of echo cancellation in full-duplex (FDX) DOCSIS and the conventional methods’ limitation.

Section 2 introduces the concept of deep learning and the principles of DNN to realize echo cancellation.

Section 3 demonstrates a proof-of-concept experiment and results analysis of DNN based echo canceller.

By clicking the "Download Paper" button, you are agreeing to our terms and conditions.

Similar Papers

Echo Cancellation Techniques for Supporting Full Duplex DOCSIS
By Hang Jin & John Chapman, Cisco Systems
2017
Full Duplex DOCSIS PHY Layer Design and Analysis for the Fiber Deep Architecture
By Richard S. Prodan, Ph.D., Broadcom Limited
2017
Full Duplex DOCSIS
By John T. Chapman and Hang Jin, Cisco
2016
Full Duplex DOCSIS over Active (N+X) Cable Networks
By Werner Coomans, Ph.D., Nokia Bell Labs; Rex Coldren, Nokia/Gainspeed
2019
Practical Considerations For Full Duplex Deployments In N+x Environments
By Dr. Bill Wall, Michael Cooper & David Job, Cox Communications
2019
Full Duplex DOCSIS Technology over HFC Networks
By Belal Hamzeh, CableLabs, Inc.
2016
Covid-19 Learnings: All Roads Lead to DOCSIS 4.0 Technology
By Jeff Finkelstein, Cox Communications; Tom Cloonan, CommScope; Doug Jones, CableLabs
2020
Optimizing the 10G Transition to Full-Duplex DOCSIS 4.0
By Richard S Prodan, Ph.D., Comcast Cable
2020
Machine Learning Techniques for Equalizing Nonlinear Distortion
By Rob Thompson, Comcast; Xiaohua Li, State University of New York at Binghamton
2020
FDX DOCSIS Line Extender: Deploying FDX DOCSIS Beyond N+0
By John T Chapman & Hang Jin, Cisco Systems
2018
More Results >>