Wednesday 09 April 2025
The quest for a seamless full-duplex wireless communication system has been a longstanding challenge in the field of telecommunications. In-band full-duplex radios, which allow devices to transmit and receive data simultaneously on the same frequency band, hold great promise for increasing network capacity and reducing latency. However, this technology is hindered by the presence of self-interference, where the transmitted signal overlaps with the received signal.
To mitigate this issue, researchers have turned to digital-aided analog self-interference cancellation (DAASIC) techniques. These methods involve using a neural network model to predict the received signal and then subtracting it from the actual received signal in order to reduce interference. However, DAASIC is limited by its reliance on accurate modeling of the transmission channel and the presence of analog-to-digital converters (ADCs), which can introduce quantization noise and saturation errors.
A new study has addressed these limitations by exploring different strategies for coping with ADC effects during DAASIC learning. The researchers simulated a digitally-aided analog self-interference cancellation system, where a neural network model is used to predict the received signal and then subtract it from the actual received signal in order to reduce interference. They compared four different approaches for modeling the ADC: backpropagation through the ADC (BPAD), straight-through estimation (STE), automatic gain control (AGC), and digital training-based approach (DTA).
The results showed that all four approaches were able to achieve similar levels of self-interference cancellation, with BPAD and STE performing slightly better than AGC and DTA. However, BPAD was found to be sensitive to the LNA gain, while STE was more robust and performed well even at higher LNA gains. The researchers also found that using an AGC system during model learning improved performance by adjusting the signal level to avoid ADC saturation.
The study highlights the importance of considering ADC effects in DAASIC systems and suggests that a combination of BPAD and AGC may be the most effective approach for achieving high-performance self-interference cancellation. The results have significant implications for the development of full-duplex wireless communication systems, which could revolutionize the way we communicate.
The researchers’ findings demonstrate that by carefully considering the ADC effects during DAASIC learning, it is possible to achieve high levels of self-interference cancellation and improve the overall performance of full-duplex wireless communication systems.
Cite this article: “Unlocking the Power of Analog Self-Interference Cancellation: A Neural Network Approach”, The Science Archive, 2025.
Digital-Aided Analog Self-Interference Cancellation, Daasic, Neural Network, Adc Effects, Full-Duplex Wireless Communication, Self-Interference Cancellation, Backpropagation, Straight-Through Estimation, Automatic Gain Control, Digital Training-Based Approach.
