Abstract:In multiple-input multiple-output (MIMO) orthogonal frequency division multiplexing (OFDM) systems, representing the whole channel only based on partial subchannels will significantly reduce the channel acquisition overhead. For such a channel mapping task, inspired by the intrinsic coupling across the space and frequency domains, this letter proposes to use interleaved learning with partial antenna and subcarrier characteristics to represent the whole MIMO-OFDM channel. Specifically, we design a complex-domain multilayer perceptron (MLP)-Mixer (CMixer), which utilizes two kinds of complex-domain MLP modules to learn the space and frequency characteristics respectively and then interleaves them to couple the learned properties. The complex-domain computation facilitates the learning on the complex-valued channel data, while the interleaving tightens the coupling of space and frequency domains. These two designs jointly reduce the learning burden, making the physics-inspired CMixer more effective on channel representation learning than existing data-driven approaches. Simulation shows that the proposed scheme brings 4.6~10dB gains in mapping accuracy compared to existing schemes under different settings. Besides, ablation studies show the necessity of complex-domain computation as well as the extent to which the interleaved learning matches the channel properties.
What problem does this paper attempt to address?
This paper attempts to solve the problem of how to efficiently represent the entire channel through partial sub - channel information in multiple - input multiple - output (MIMO) and orthogonal frequency - division multiplexing (OFDM) systems. Specifically, the paper focuses on how to reduce the overhead of channel acquisition, that is, by utilizing the inherent coupling characteristics in the spatial and frequency domains to infer the state of the entire MIMO - OFDM channel from the information of some antennas and sub - carriers.
### Problem Background
In MIMO and OFDM technologies, in order to achieve high - efficiency spectrum utilization, it is usually necessary to obtain channel state information (CSI) in real - time. However, as the number of antennas and sub - carriers increases, the amount of CSI data increases significantly, resulting in excessive signaling overhead. To solve this problem, the paper proposes a channel mapping method based on interleaved learning, aiming to reconstruct the entire channel through partial sub - channel information.
### Solution
The main contributions of the paper include:
1. **Revealing the inherent coupling between the spatial and frequency characteristics of the MIMO - OFDM channel**: By analyzing the channel model, it is found that there are shared large - scale characteristics in the channel responses between different sub - carriers and antennas.
2. **Proposing an interleaved learning framework**: This framework can explore this cross - domain internal correlation and improve the accuracy of channel representation by alternately learning spatial and frequency characteristics.
3. **Designing a complex - domain MLP - Mixer (CMixer) module**: This module uses two complex - domain MLP modules to learn spatial and frequency characteristics respectively, and combines these two characteristics in an interleaved manner, so as to more effectively process complex - valued channel data.
4. **Constructing the CMixer model**: By stacking multiple CMixer layers, a deep neural network model is formed for the channel mapping task.
### Experimental Results
The experimental results show that the proposed CMixer model has significant advantages over existing methods under different settings. It can improve the channel mapping accuracy by 4.6 - 10 dB, and can reduce the known channel size by up to 67.3% under the same accuracy requirements.
### Conclusion
By introducing complex - domain calculations and interleaved learning, the paper successfully improves the performance of channel mapping, proving that these designs not only rely on data fitting, but also fully utilize the physical characteristics of the channel. This provides new ideas and methods for using deep - learning techniques to solve problems in wireless communication systems in the future.
### Formula Summary
1. **Channel Model**:
\[
h(f)=\sum_{p = 1}^{P}\alpha_p e^{-j2\pi f\tau_p}a(\vec{p})
\]
where \( f \) is the carrier frequency, \( \alpha_p \) is the amplitude attenuation, \( \tau_p \) is the propagation delay, \( \vec{p} \) is the three - dimensional unit direction vector of the \( p \) - th path, and \( a(\vec{p}) \) is the array response vector.
2. **CSI Matrix**:
\[
H=[h(f_1),h(f_2),\cdots,h(f_{N_c})]
\]
where \( f_i = f_0+\Delta f_i \), \( f_0 \) is the reference frequency, and \( \Delta f_i \) is the frequency offset between the \( i \) - th sub - carrier and the reference frequency.
3. **Loss Function**:
\[
\text{Loss}(\Theta)=\frac{1}{N}\sum_{\text{num}=1}^{N}\|H_{\text{num}}-\hat{H}_{\text{num}}\|_2^2
\]
where \( \Theta \) is the set of model parameters, \( N \) is the number of channel samples in the training set, and \( \| \cdot \|_2 \) is the Euclidean norm.
4. **Normalized Mean - Square Error (NMSE)**:
\[
\text{NMSE}=E\left[\frac{\|H - \hat{H}\|_2^2}{\|H\|_2^2}\right]