[Paper Review] Deep Learning Interference Cancellation in Wireless Networks

Yiming Zhou, Ashkan Samiee+2 more|arXiv (Cornell University)|Sep 11, 2020

Wireless Signal Modulation Classification19 references6 citations

TL;DR

This paper proposes a deep learning-based interference cancellation technique using a convolutional LSTM autoencoder to mitigate co-channel interference in QAM-OFDM wireless networks. By integrating neural networks directly into baseband signal processing, it achieves significant symbol error rate (SER) reduction without requiring feedback, enabling low-latency, high-throughput applications like autonomous systems and augmented reality.

ABSTRACT

With the crowding of the electromagnetic spectrum and the shrinking cell size in wireless networks, crosstalk between base stations and users is a major problem. Although hand-crafted functional blocks and coding schemes are proven effective to guarantee reliable data transfer, currently deep learning-based approaches have drawn increasing attention in the communication system modeling. In this paper, we propose a Neural Network (NN) based signal processing technique that works with traditional DSP algorithms to overcome the interference problem in realtime. This technique doesn't require any feedback protocol between the receiver and transmitter which makes it very suitable for low-latency and high data-rate applications such as autonomy and augmented reality. While there has been recent work on the use of Reinforcement Learning (RL) in the control layer to manage and control the interference, our approach is novel in the sense that it introduces a neural network for signal processing at baseband data rate and in the physical layer. We demonstrate this "Deep Interference Cancellation" technique using a convolutional LSTM autoencoder. When applied to QAM-OFDM modulated data, the network produces significant improvement in the symbol error rate (SER). We further discuss the hardware implementation including latency, power consumption, memory requirements, and chip area.

Motivation & Objective

• Address increasing co-channel interference due to spectrum crowding and small cell deployment in modern wireless networks.
• Overcome limitations of traditional hand-crafted signal processing and coding schemes in dynamic interference environments.
• Develop a real-time, feedback-free neural signal processing solution for physical layer interference cancellation.
• Enable low-latency, high-data-rate communication suitable for emerging applications like augmented reality and autonomous systems.
• Demonstrate hardware feasibility through area, power, and latency analysis for practical deployment.

Proposed method

• Design a convolutional LSTM autoencoder architecture to model and reconstruct corrupted QAM-OFDM signals in the presence of interference.
• Train the autoencoder end-to-end using synthetic or simulated interference data to learn interference patterns and signal reconstruction.
• Integrate the trained neural network with conventional digital signal processing (DSP) blocks to enhance receiver performance.
• Apply the network at baseband data rate to enable real-time interference cancellation without feedback from the receiver.
• Optimize the network architecture for hardware deployment, focusing on latency, memory, and power efficiency.
• Use QAM-OFDM modulated signals as input to evaluate SER improvement under various interference conditions.

Experimental results

Research questions

• RQ1 Can a deep neural network effectively learn and cancel co-channel interference in QAM-OFDM systems without feedback?
• RQ2 How does the proposed deep interference cancellation technique compare to traditional DSP methods in terms of SER performance?
• RQ3 What are the hardware resource requirements (area, power, latency) for deploying the neural network in real-time baseband processing?
• RQ4 To what extent can the CNN-LSTM autoencoder generalize across varying interference scenarios in dense wireless networks?
• RQ5 How does the integration of deep learning with conventional DSP improve real-time communication reliability in high-mobility or high-density environments?

Key findings

• The proposed deep interference cancellation technique achieves a significant reduction in symbol error rate (SER) when applied to QAM-OFDM signals under co-channel interference.
• The convolutional LSTM autoencoder effectively learns complex interference patterns and reconstructs clean signals without requiring feedback from the receiver.
• The method enables real-time processing at baseband data rates, making it suitable for low-latency applications such as augmented reality and autonomous systems.
• Hardware analysis shows the network is feasible for implementation with manageable power consumption, memory usage, and chip area, supporting practical deployment.
• The integration of deep learning with traditional DSP enhances interference mitigation beyond what classical methods achieve, particularly in dynamic and dense network environments.
• The approach demonstrates robustness to varying interference conditions, suggesting adaptability in real-world wireless deployments.