Explainable AI for Enhancing Efficiency of DL-based Channel Estimation

TL;DR

This work addresses the lack of interpretability in DL-based channel estimation for wireless links by introducing XAI-CHEST, a perturbation-based, model-agnostic pre-model framework that identifies a subset of input subcarriers deemed relevant to the decision of a baseline estimator U. It formalizes the interpretability mechanism with losses $L_N$ and $L_U$, along with an interpretability penalty $L_X$, and derives a noise-threshold tuning approach to select the optimal input subset $\Psi$ that preserves BER while reducing DL complexity. The key contributions are the theoretical foundations for the XAI-CHEST loss functions, analytical noise-threshold tuning, demonstration that restricting inputs to the identified relevant set improves BER and enables architecture optimization (e.g., shallower FNNs with fewer FLOPS). Simulation results across LFS/HFS channels, various modulations, RF nonlinearity, and training SNR show substantial BER gains and up to multi-fold reductions in computational complexity, yielding more trustworthy and efficient DL-based channel estimation for next-generation wireless systems.

Abstract

The support of artificial intelligence (AI) based decision-making is a key element in future 6G networks, where the concept of native AI will be introduced. Moreover, AI is widely employed in different critical applications such as autonomous driving and medical diagnosis. In such applications, using AI as black-box models is risky and challenging. Hence, it is crucial to understand and trust the decisions taken by these models. Tackling this issue can be achieved by developing explainable AI (XAI) schemes that aim to explain the logic behind the black-box model behavior, and thus, ensure its efficient and safe deployment. Recently, we proposed a novel perturbation-based XAI-CHEST framework that is oriented toward channel estimation in wireless communications. The core idea of the XAI-CHEST framework is to identify the relevant model inputs by inducing high noise on the irrelevant ones. This manuscript provides the detailed theoretical foundations of the XAI-CHEST framework. In particular, we derive the analytical expressions of the XAI-CHEST loss functions and the noise threshold fine-tuning optimization problem. Hence the designed XAI-CHEST delivers a smart input feature selection methodology that can further improve the overall performance while optimizing the architecture of the employed model. Simulation results show that the XAI-CHEST framework provides valid interpretations, where it offers an improved bit error rate performance while reducing the required computational complexity in comparison to the classical DL-based channel estimation.

Figures (13)

• Figure 1: XAI categories, concepts, and taxonomy.
• Figure 2: OFDM transmitter-receiver block diagram.
• Figure 3: Block diagram of the XAI-CHEST framework. The first step is to train the $U$ model and freeze its parameters $\theta_U$. After that, the N model is trained where the objective is to boost the growth of $\bm{b} _{\Phi_{i}}^{\prime}$ while preserving the same performance of the pre-trained $U$ model \ref{['eq:lossn']}. Finally, the subcarriers are filtered based on $\bm{b} _{\Phi_{i}}^{\prime}$, where higher noise weight signifies that the corresponding subcarrier is irrelevant (red colors). In contrast, low noise weights mean that the subcarriers are relevant (blue colors). We recall that in this work we are considering the STA-FNN channel estimation scheme, hence, $\Phi = \text{STA}$.
• Figure 4: Normalized restricted loss function.
• Figure 5: Fine-tuning of the noise threshold $\gamma$ considering the HFS channel model and QPSK modulation.

Theorems & Definitions (1)

• Lemma 1