Abstract
Channel estimation and signal detection are very challenging for an orthogonal frequency division multiplexing (OFDM) system without cyclic prefix (CP). In this article, deep learning based on orthogonal approximate message passing (DL-OAMP) is used to address these problems. The DL-OAMP receiver includes a channel estimation neural network (CE-Net) and a signal detection neural network based on OAM-P, called OAMP-Net. The CE-Net is initialized by the least square channel estimation algorithm and refined by minimum mean-squared error (MMSE) neural network. The OAMP-Net is established by unfolding the iterative OAMP algorithm and adding some trainable parameters to improve the detection performance. The DL-OAMP receiver is with low complexity and can estimate time-varying channels with only a single training. Simulation results demonstrate that the bit-error rate (BER) of the proposed scheme is lower than those of competitive algorithms for high-order modulation.
Introduction
Orthogonal Frequency Division Multiplexing (OFDM) is a cornerstone of modern wireless communication, utilized in 4G LTE and Wi-Fi. Traditionally, OFDM relies on a Cyclic Prefix (CP) to mitigate Inter-Symbol Interference (ISI) and Inter-Carrier Interference (ICI). However, the CP introduces significant overhead, often consuming up to 25% of bandwidth and energy. Removing the CP would enhance spectral efficiency but introduces complex interference that conventional receivers struggle to manage.
This paper introduces DL-OAMP, a model-driven deep learning receiver specifically designed for CP-free OFDM systems. By integrating the physical intuition of the Orthogonal Approximate Message Passing (OAMP) algorithm with the adaptive capabilities of deep learning, the proposed receiver achieves superior performance compared to traditional interference cancellation and purely data-driven neural networks.
Architecture
The DL-OAMP receiver consists of two primary components: the Channel Estimation Network (CE-Net) and the Signal Detection Network (OAMP-Net).
CE-Net (Channel Estimation)
CE-Net adopts a model-driven approach. It begins with a standard Least Square (LS) estimate, which is then refined by a Minimum Mean-Squared Error (MMSE) neural network. This initialization allows the network to refine channel estimates with very little training data compared to generic architectures.
OAMP-Net (Signal Detection)
The core of the receiver is OAMP-Net, which "unfolds" the iterative OAMP algorithm into a neural network structure. Each layer in the network corresponds to one iteration of the OAMP algorithm. Unlike the standard algorithm, OAMP-Net introduces learnable scalar parameters at each layer. This design ensures the network remains lightweight—requiring only a few trainable parameters—and computationally efficient, as the parameter count is independent of the number of subcarriers.
Performance Evaluation
The system was evaluated using simulated data based on the WINNER II channel model, reflecting realistic multi-path fading environments. The performance was tested across various modulation schemes, including 16QAM and 64QAM.
Key findings from the simulations include:
- Superior BER: In 64QAM scenarios, DL-OAMP significantly outperformed existing methods such as RISIC (Residual ISI Cancellation), ComNet, and standard OAMP.
- Efficiency Gains: The DL-OAMP receiver nearly matched the performance of ideal systems that utilize a Cyclic Prefix (the "CSI_ML_CP" benchmark), suggesting that the bandwidth lost to CP can be reclaimed without substantial signal degradation.
- Adaptability: The model demonstrates the ability to adapt to time-varying channels with only a single training session, highlighting its robustness for real-world applications.
Conclusion
The DL-OAMP receiver offers a promising path toward high-efficiency, CP-free OFDM systems. By leveraging model-driven deep learning, it achieves high performance with low computational complexity. Future work will focus on reducing pilot dependency through blind estimation techniques and extending the architecture to Massive MIMO systems for 5G and 6G applications. While the sequential nature of the unfolded layers may present hardware latency challenges, the drastic reduction in parameter count makes it a strong candidate for real-time implementation.
Summary updated