Deep Convolutional Neural Networks for Link Adaptations in MIMO-OFDM Wireless Systems

This letter proposes a deep convolutional neural network (DCNN) approach for adaptive modulation and coding in practical multiple-input, multiple-output orthogonal frequency division multiplexing (MIMO-OFDM) systems. Our target is to maximize the throughput and fulfill a packet error rate constraint. We consider practical impairments of MIMO-OFDM receiver, such as imperfect timing synchronization, carrier frequency offset correction, and channel estimation. We treat the estimated channel state information and the noise standard deviation as input features to the DCNN. The main advantages of the proposed approach are: 1) it learns the characteristics of the MIMO-OFDM channel properly and predicts the suitable modulation and coding scheme and 2) it does not need complex features selection.

I. Introduction

The spectrum efficiency of a wireless link predicate on its ability to conform with the varying wireless channel. To maximize the throughput under packet error rate (PER) constraint, multiple-input, multiple-output orthogonal frequency division multiplexing (MIMO-OFDM) link requires adaptive modulation and coding (AMC) by selecting the modulation and coding scheme (MCS) based on the channel state information (CSI) [1]. Considering the impacts of OFDM modulation, MIMO processing, error control coding, bit interleaving and frequency-selective channel, modeling of MIMO-OFDM systems is challenging. Thus, the previous AMC approaches assume approximations to simplify the system models (e.g., [2]). However, these simplified models cannot reflect all the system factors and the associated AMC approaches is impacted [1].