Turbo-ICL: In-Context Learning-Based Turbo Equalization

TL;DR

The paper addresses reliable data recovery in quantized MIMO with turbo decoding without explicit channel state information. It introduces an in-context learning-based soft equalizer that ingests pilot examples and decoder priors through a turbo-compatible prompt, producing posterior symbol distributions rather than hard decisions. Two backbones, Transformer and state-space models, are developed and pre-trained to generalize across channels, quantization levels, and SNRs, showing superior robustness to nonlinear distortions and higher-order modulations compared to model-based baselines, even with perfect CSI in some regimes. The approach offers a CSI-free, adaptive receiver design with demonstrated efficiency and scalability, enabling practical deployment in challenging wireless scenarios.

Abstract

This paper introduces a novel in-context learning (ICL) framework, inspired by large language models (LLMs), for soft-input soft-output channel equalization in coded multiple-input multiple-output (MIMO) systems. The proposed approach learns to infer posterior symbol distributions directly from a prompt of pilot signals and decoder feedback. A key innovation is the use of prompt augmentation to incorporate extrinsic information from the decoder output as additional context, enabling the ICL model to refine its symbol estimates iteratively across turbo decoding iterations. Two model variants, based on Transformer and state-space architectures, are developed and evaluated. Extensive simulations demonstrate that, when traditional linear assumptions break down, e.g., in the presence of low-resolution quantization, ICL equalizers consistently outperform conventional model-based baselines, even when the latter are provided with perfect channel state information. Results also highlight the advantage of Transformer-based models under limited training diversity, as well as the efficiency of state-space models in resource-constrained scenarios.

Figures (12)

• Figure 1: (a) Example of a prompt and the corresponding output of a large language model (LLM) performing in-context learning (ICL). Given a few input-output examples of a task (e.g., word reversal), the model generalizes the underlying pattern and produces the correct output for a new query. (b)-(c) The same principle can be applied to the task of channel equalization. The ICL-based equalizer is prompted with a set of channel input-output pairs, corresponding to pilots, along with a received data symbol, and it returns the estimated transmitted symbol.
• Figure 2: Block diagrams of coded MIMO transmission with a quantized front end at the receiver.
• Figure 3: Block diagrams of turbo equalization (a) with a conventional soft channel equalizer assisted by estimated CSI, and (b) with an ICL soft channel equalizer without explicit CSI estimation.
• Figure 4: Structure of an ICL-based soft equalizer.
• Figure 5: Illustration of an arbitrary layer in the (a) Transformer (b) SSM model.