[Paper Review] CNN aided Weighted Interpolation for Channel Estimation in Vehicular Communications

Abdul Karim Gizzini, Marwa Chafii+3 more|arXiv (Cornell University)|Apr 18, 2021

Millimeter-Wave Propagation and Modeling36 references39 citations

TL;DR

This paper proposes a novel CNN-aided weighted interpolation (WI) channel estimator for IEEE 802.11p vehicular communications that inserts adaptive pilot symbols into the frame to enable low-latency, high-data-rate, and robust channel estimation in high-mobility scenarios. By leveraging time-domain weighted interpolation and optimized super-resolution or denoising CNNs, the method achieves significant performance gains over prior deep learning estimators with up to 70× lower computational complexity than ChannelNet and 7,027× lower than 2D LMMSE, while improving transmission data rate and reducing buffering latency.

ABSTRACT

IEEE 802.11p standard defines wireless technology protocols that enable vehicular transportation and manage traffic efficiency. A major challenge in the development of this technology is ensuring communication reliability in highly dynamic vehicular environments, where the wireless communication channels are doubly selective, thus making channel estimation and tracking a relevant problem to investigate. In this paper, a novel deep learning (DL)-based weighted interpolation estimator is proposed to accurately estimate vehicular channels especially in high mobility scenarios. The proposed estimator is based on modifying the pilot allocation of the IEEE 802.11p standard so that more transmission data rates are achieved. Extensive numerical experiments demonstrate that the developed estimator significantly outperforms the recently proposed DL-based frame-by-frame estimators in different vehicular scenarios, while substantially reducing the overall computational complexity.

Motivation & Objective

• Address the challenge of accurate and low-complexity channel estimation in high-mobility vehicular environments with doubly selective fading.
• Overcome the limitations of existing frame-by-frame deep learning estimators like ChannelNet and TS-ChannelNet, which suffer from high computational complexity, performance degradation at high mobility, and high receiver buffering latency.
• Improve transmission data rate by fully allocating OFDM symbols to data subcarriers while inserting only a minimal number of pilot symbols at strategic positions.
• Reduce receiver latency by enabling subframe-based channel estimation instead of waiting for the full frame.
• Develop a hybrid, adaptive, and robust estimator that dynamically adjusts pilot insertion based on mobility conditions to maintain performance across diverse vehicular scenarios.

Proposed method

• Proposes a novel pilot allocation scheme that inserts one or more full-pilot OFDM symbols at the end of the frame, with the number of pilots increasing with mobility.
• Applies weighted interpolation in the time domain using only the inserted pilot symbols to estimate the channel for all data symbols, avoiding frequency-domain interpolation.
• Introduces two variants: WI-FP-ALS (using weighted alternating least squares) and WI-FP-SLS (using simple least squares) for channel estimation.
• Integrates optimized super-resolution CNN (SR-CNN) or denoising CNN (DN-CNN) as post-processing modules to enhance estimation accuracy.
• Adapts the frame structure and pilot count based on mobility: one pilot symbol in low mobility, increasing to two or three in high and very high mobility.
• Processes the frame in subframes, allowing channel estimation to begin immediately after each subframe is received, significantly reducing buffering time.

Experimental results

Research questions

• RQ1 Can a deep learning-based channel estimator achieve superior performance to existing frame-by-frame estimators like ChannelNet and TS-ChannelNet in high-mobility vehicular scenarios?
• RQ2 Can a modified IEEE 802.11p frame structure with adaptive pilot insertion reduce computational complexity while maintaining or improving estimation accuracy?
• RQ3 To what extent can weighted time-domain interpolation outperform conventional 2D interpolation in doubly selective channels?
• RQ4 How does subframe-based processing reduce receiver buffering latency compared to full-frame processing in real-time vehicular applications?
• RQ5 Can optimized CNN architectures (SR-CNN, DN-CNN) be effectively integrated into a lightweight interpolation framework to enhance performance without increasing complexity?

Key findings

• The proposed WI-FP-ALS-SR-CNN estimator achieves 70 times lower computational complexity than ChannelNet and 39 times lower than TS-ChannelNet.
• The proposed estimators reduce computational complexity by at least 7,027.35 times compared to the conventional 2D LMMSE estimator while maintaining acceptable bit error rate (BER) performance.
• The WI-2P and WI-3P estimators achieve 8.08% and 7.83% higher transmission data rate (TDR) than the standard IEEE 802.11p, respectively, due to full data subcarrier allocation.
• The WI-2P and WI-3P estimators reduce buffering time to 400 µs and 265 µs, respectively, compared to 800 µs for ChannelNet and TS-ChannelNet, enabling earlier channel estimation.
• The FP-ALS-DN-CNN variant, used in high and very high mobility, is 12 times more complex than FP-ALS-SR-CNN but still significantly less complex than ChannelNet and TS-ChannelNet.
• Simulation results confirm that the proposed estimators outperform ChannelNet and TS-ChannelNet across all mobility scenarios, especially in high-mobility conditions where prior methods degrade significantly.