[Paper Review] Power of Deep Learning for Channel Estimation and Signal Detection in OFDM Systems

Hao Ye, Geoffrey Ye Li+1 more|arXiv (Cornell University)|Aug 28, 2017

Wireless Signal Modulation Classification6 references76 citations

TL;DR

The paper proposes an end-to-end deep learning approach for joint channel estimation and symbol detection in OFDM, trained offline and deployed online to recover data without explicit CSI estimation, showing competitive performance and robustness under distortion and limited pilots.

ABSTRACT

This article presents our initial results in deep learning for channel estimation and signal detection in orthogonal frequency-division multiplexing (OFDM). OFDM has been widely adopted in wireless broadband communications to combat frequency-selective fading in wireless channels. In this article, we take advantage of deep learning in handling wireless OFDM channels in an end-to-end approach. Different from existing OFDM receivers that first estimate CSI explicitly and then detect/recover the transmitted symbols with the estimated CSI, our deep learning based approach estimates CSI implicitly and recovers the transmitted symbols directly. To address channel distortion, a deep learning model is first trained offline using the data generated from the simulation based on the channel statistics and then used for recovering the online transmitted data directly. From our simulation results, the deep learning based approach has the ability to address channel distortions and detect the transmitted symbols with performance comparable to minimum mean-square error (MMSE) estimator. Furthermore, the deep learning based approach is more robust than conventional methods when fewer training pilots are used, the cyclic prefix (CP) is omitted, and nonlinear clipping noise is presented. In summary, deep learning is a promising tool for channel estimation and signal detection in wireless communications with complicated channel distortions and interferences.

Motivation & Objective

• Motivate deep learning as an end-to-end tool for OFDM channel estimation and data detection.
• Demonstrate that a DNN can implicitly learn channel characteristics and recover transmitted symbols without explicit CSI.
• Show robustness of the DL approach under reduced pilots, CP omission, and nonlinear clipping noise.
• Compare DL performance with LS and MMSE under various channel/distortion conditions.

Proposed method

• Train a multi-layer DNN offline using simulated OFDM frames with pilots and data under varied channel conditions.
• Input the received OFDM blocks (pilot + data) to the DNN to jointly estimate transmitted data without explicit CSI.
• Use L2 loss between predicted and transmitted data to optimize network weights.
• Implement an OFDM system with 64 subcarriers and CP length 16, using QPSK, under WINNER II urban channel models.
• Evaluate performance via BER across SNRs and pilot counts, comparing with LS and MMSE."],
• research_questions([
• Can a DNN jointly estimate channel effects and recover OFDM symbols without explicit CSI?
• How does the DL approach compare to LS and MMSE under varying pilots, CP presence/absence, and clipping noise?
• Is the DL model robust to mismatches between training and deployment channel statistics?
• What is the impact of nonlinear distortion on DL versus traditional estimators?

Experimental results

Research questions

• RQ1 Can a DNN jointly estimate channel effects and recover OFDM symbols without explicit CSI?
• RQ2 How does the DL approach compare to LS and MMSE under varying pilots, CP presence/absence, and clipping noise?
• RQ3 Is the DL model robust to mismatches between training and deployment channel statistics?
• RQ4 What is the impact of nonlinear distortion on DL versus traditional estimators?

Key findings

• DL-based joint estimation and detection achieves BER comparable to MMSE when enough pilots are available.
• With fewer pilots (e.g., 8), the DL method maintains BER improvements over LS and approaches MMSE performance.
• DL exhibits robustness when CP is omitted, outperforming LS/MMSE under such distortion.
• DL is more robust to clipping noise compared to MMSE, especially at higher SNRs.
• Mismatch in training/deployment channel statistics does not severely degrade DL symbol detection performance.