5G+AI has become the most popular topic for the research of intelligent wireless communications in recent years. As such, Samsung R&D Institute China - Beijing (SRC-B) continues to trailblaze in the field of AI + wireless communication. Following the granting of international standardization and industrialization recognition to Samsung, two teams from SRC-B stood out from over 300 worldwide teams and made formidable achievements in the 2nd Wireless Communication AI Competition (WAIC), a prestigious wireless + AI competition attracting global participants. The excellent results manifest how SRC-B has a strong capability in AI-based channel estimation fields and an unstoppable spirit of innovation.
 

The 2nd Wireless Communication AI Competition (WAIC)
WAIC is hosted by the 5G+AI Work Group of IMT-2020 (5G) Promotion Group, which is committed to conducting research on the requirements, theories, and technologies of deep integration of 5G and AI, and promoting the international standardization and industrialization process of 5G and AI integration. More than 300 teams from operators, vendors, and universities joined the contest globally.
The 2nd WAIC was held online in July and August 2021, consisting of two tracks of AI-Based Channel State Information Feedback and AI-Based Channel Estimation. These two promising techniques for air interface ignited heated discussion in present academics and B5G/6G standardization organizations.
SRC-B mainly focused on the track of channel estimation, which is the basic module to ensure the performance of wireless communication systems. 

AI-Based Channel Estimation Track
Accurate channel estimation (CE) and effective channel state information (CSI) feedback are the basic conditions to ensure the performance of wireless communication systems. The task of the session is to estimate the channel information on all resource elements based on a demodulation reference signal (DMRS) on a small number of resource elements. The challenge consisted of two subtasks, namely, time and frequency channel recovery and generalization performance of the AI model in different channel conditions. 
Two SRC-B teams participated in the 2nd WAIC in the AI-Based Channel Estimation track.

A Simple Mixer-MLP Based on SNR Feedback
The participants need to consider making use of the nonlinear recovery performance of AI and the reference signals of a very small number to estimate the channel information on the whole channel, which could be used for the signal demodulation of the receiver. The SRC-B team developed a simple Mixer-Multilayer Perceptron (MLP) based on signal¬–noise ratio (SNR) feedback solution and ultimately took first place out of more than 300 teams.
 

△ SRC-B Team 1: Advanced Research and Standard Team & AI Lab
Competing with the world’s top-tier universities, companies, and research labs, SRC-B Advanced Research & Standard Team collaborated with SRC-B AI-Lab, and made a breakthrough in wireless channel estimation issue. Based on refined Mixer model architecture, the channel information is de-noised and recovered, which achieves astonishing accuracy improvement on this task and gets a much higher score compared with existing algorithms. 
In addition to the innovative Mixer Block as the Network backbone, the SRC-B team also introduced approaches to adapt the machine learning model to the task of channel estimation. SNR information will feedback on loss function, which will improve the model extract information ability. Meanwhile, SNR is also input in a neural network with a normalization matrix to suggest a model. The simple Mixer-MLP based on SNR feedback provides more interaction between the time domain and the frequency domain to improve model performance, which suggests a brand-new way toward AI-aided wireless data processing

Noise Amplitude Weighted MLP-Mixer 
By using the AI’s nonlinear recovery performance, the accuracy of the signal demodulation of the receiver is improved compared to the traditional linear algorithms, such as Wiener filtering.
Another team from SRC-B proposed a noise amplitude weighted MLP-mixer solution. By using pilot position information and artificial interpolation, full channel information was recovered. MLP-mixer network was used to extract time-frequency information between and within different groups. Moreover, the SNR information was applied to the amplitude weighted coefficient of network jump connection, which further improved the generalization of different SNRs.
 

△ SRC-B Team 2: NW Tech Team
The AI-enabled wireless communication system can enhance the performance of the conventional algorithm. The idea of MLP-mixer solution for channel estimation is to apply time-frequency 2D correlation of fading channel to recover the whole channel state with limited pilot information. The results show that the performance of the AI-based scheme is highly competitive with the traditional scheme. This idea also can be used in other problems of the Orthogonal Frequency Division Multiplexing (OFDM) system to enhance performance. Finally, the SRC-B team won third place with a score of 41.89. 

Channel estimation is a key technique to wireless communication systems and has long been studied for a long time. In the future, SRC-B will continue moving forward on the unprecedented way of emerging 5G+AI technology to bring high-quality solutions and algorithms to promote technical development.