How Critical is Site-Specific RAN Optimization? 5G Open-RAN Uplink Air Interface Performance Test and Optimization from Macro-Cell CIR Data

Johnathan Corgan,Nitin Nair,Rajib Bhattacharjea,Wan Liu,Serhat Tadik,Tom Tsou,Timothy J. O'Shea

2024-10-25

Abstract:In this paper, we consider the importance of channel measurement data from specific sites and its impact on air interface optimization and test. Currently, a range of statistical channel models including 3GPP 38.901 tapped delay line (TDL), clustered delay line (CDL), urban microcells (UMi) and urban macrocells (UMa) type channels are widely used for air interface performance testing and simulation. However, there remains a gap in the realism of these models for air interface testing and optimization when compared with real world measurement based channels. To address this gap, we compare the performance impacts of training neural receivers with 1) statistical 3GPP TDL models, and 2) measured macro-cell channel impulse response (CIR) data. We leverage our OmniPHY-5G neural receiver for NR PUSCH uplink simulation, with a training procedure that uses statistical TDL channel models for pre-training, and fine-tuning based on measured site specific MIMO CIR data. The proposed fine-tuning method achieves a 10% block error rate (BLER) at a 1.85 dB lower signal-to-noise ratio (SNR) compared to pre-training only on simulated TDL channels, illustrating a rough magnitude of the gap that can be closed by site-specific training, and gives the first answer to the question "how much can fine-tuning the RAN for site-specific channels help?"

Signal Processing,Machine Learning

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **Limitations of statistical channel models in performance testing and optimization of actual wireless communication systems**. Specifically, the gap between currently widely - used statistical channel models (such as 3GPP 38.901 TDL, CDL, UMi, and UMa, etc.) and real - world measurement data has led to inaccuracies in wireless interface performance testing and optimization.

### Specific manifestations of the problem:
1. **Limitations of statistical models**: Although existing statistical channel models can capture some dynamic characteristics of wireless channels, they cannot fully simulate complex situations in the real environment, such as the geometric structure of specific locations, multipath effects, interference, etc.
2. **Performance gap**: Algorithms designed and tested based on statistical models may encounter performance degradation problems in practical applications because these models do not take into account all real - world factors.

### Goals of the paper:
To bridge this gap, the paper evaluates the performance impact by comparing two methods of training neural receivers:
- Pretraining using the statistical 3GPP TDL model.
- Fine - tuning using measured macro - cell channel impulse response (CIR) data.

### Main findings:
Through experiments, the paper shows that fine - tuning using measured CIR data can significantly improve the performance of neural receivers. Specifically, under the condition of 10% block error rate (BLER), the fine - tuned model can operate at a signal - to - noise ratio (SNR) 1.85 dB lower than the model pretrained only with the TDL model. This indicates that fine - tuning for specific sites can significantly improve the performance of wireless communication systems.

### Formula representation:
- **Signal - to - noise ratio (SNR)**: \[ \text{SNR}=\frac{P_{\text{signal}}}{P_{\text{noise}}} \]
- **Block error rate (BLER)**: \[ \text{BLER}=\frac{\text{Number of erroneous blocks}}{\text{Total number of blocks}} \]

### Conclusion:
This research answers for the first time the question of "How much help can optimizing RAN for specific sites bring?" and shows that by combining measured data for fine - tuning, the performance of wireless communication systems can be significantly improved. This is of great significance for the design and optimization of future 5G and higher - level communication systems.