Unsupervised Learning of Adaptive Codebooks for Deep Feedback Encoding in FDD Systems

Turan, Nurettin; Koller, Michael; Bazzi, Samer; Xu, Wen; Utschick, Wolfgang

doi:10.1109/ieeeconf53345.2021.9723406

Cited by 9 publications

(14 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In [3], the observation has been made in the context of training autoencoders. Similar observations have since been made in [4] for DL channel estimation and in [5], [6] for codebook design. In this work, we also utilize the idea to centrally learn DL-related functionalities at the BS using UL training data.…”

Section: Introductionsupporting

confidence: 74%

See 1 more Smart Citation

GMM-based Codebook Construction and Feedback Encoding in FDD Systems

Turan¹,

Koller²,

Benedikt³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

We propose a precoder codebook construction and feedback encoding scheme which is based on Gaussian mixture models (GMMs). In an offline phase, the base station (BS) first fits a GMM to uplink (UL) training samples. Thereafter, it designs a codebook in an unsupervised manner by exploiting the GMM's clustering capability. We design one codebook entry per GMM component. After offloading the GMM-but not the codebook-to the mobile terminal (MT) in the online phase, the MT utilizes the GMM to determine the best fitting codebook entry. To this end, no channel estimation is necessary at the MT. Instead, the MT's observed signal is used to evaluate how responsible each component of the GMM is for the signal. The feedback consists of the index of the GMM component with the highest responsibility and the BS then employs the corresponding codebook entry. Simulation results show that the proposed codebook design and feedback encoding scheme outperforms conventional Lloyd clustering based codebook design algorithms, especially in configurations with reduced pilot overhead.

show abstract

Section: Introductionsupporting

confidence: 74%

“…This optimization problem is solved via projected gradient descent (PGD), cf. [5], [21]. In summary, the GMM is used twice: Once for codebook construction (done offline) and thereafter to determine a feedback index (done online).…”

Section: A Proposed Codebook Construction and Encoding Schemementioning

confidence: 99%

GMM-based Codebook Construction and Feedback Encoding in FDD Systems

Turan¹,

Koller²,

Benedikt³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…Consequently, for a certain propagation scenario it can be expected that there exist different pairs of antenna positions and carrier frequencies with identical CSI [11]. In [13], this UL-DL conjecture is exploited for adaptive codebook construction and feedback generation.…”

Section: A Proposed Learning Strategymentioning

confidence: 99%

Centralized Learning of the Distributed Downlink Channel Estimators in FDD Systems using Uplink Data

Benedikt¹,

Turan²,

Koller³

et al. 2021

Preprint

View full text Add to dashboard Cite

In this work, we propose a convolutional neural network (CNN) based low-complexity approach for downlink (DL) channel estimation (CE) in frequency division duplex (FDD) systems. In contrast to existing work, we use training data which solely stems from the uplink (UL) domain. This allows to learn the CNN centralized at the base station (BS). After training, the network parameters are offloaded to mobile terminals (MTs) within the coverage area of the BS. The MTs can then obtain channel state information (CSI) of the MIMO channels with the low-complexity CNN estimator. This circumvents the necessity of an infeasible amount of feedback, i.e., acquisition of training data at the user, and the offline training phase at each MT. Numerical results show that the CNN which is trained solely based on UL data performs equally well as the network trained based on DL data. Furthermore, the approach is able to outperform state-ofthe-art CE algorithms.

show abstract

“…Similarly, in [31], the channel is compressed at the UE based on compressive sensing and then recovered through a DL architecture at the BS. In [32], a NN has been proposed to directly map noisy pilot observations to their optimal feedback index. In this way, neither channel estimation nor knowledge of the feedback codebook are necessary at the UE.…”

Section: Introductionmentioning

confidence: 99%

“…First, very large training datasets are tipically required: e.g., hundreds of thousands of downlink training April 12, 2022 DRAFT samples are used in many of the aforementioned studies [9], [11], [16], [17], [19], [23], [26], [27], [31], [33], [35]. This problem has been recently addressed by training the DL architectures solely with uplink channel samples, assumed available at the BS [30], [18], [32], [36]. These works exploit the conjecture that learning at the uplink frequency can be transferred at the downlink frequency with no further modification, as proposed for the first time in [30] and validated in [36].…”

Section: Introductionmentioning

confidence: 99%

Machine Learning-Based CSI Feedback With Variable Length in FDD Massive MIMO

Nerini¹,

Rizzello²,

Joham³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

To fully unlock the benefits of multiple-input multiple-output (MIMO) networks, downlink channel state information (CSI) is required at the base station (BS). In frequency division duplex (FDD) systems, the CSI is acquired through a feedback signal from the user equipment (UE). However, this may lead to an important overhead in FDD massive MIMO systems. Focusing on these systems, in this study, we propose a novel strategy to design the CSI feedback. Our strategy allows to optimally design the feedback with variable length, while reducing the parameter number at the UE. Specifically, principal component analysis (PCA) is used to compress the channel into a latent space with adaptive dimensionality. To quantize this compressed channel, the feedback bits are smartly allocated to the latent space dimensions by minimizing the normalized mean squared error (NMSE) distortion. Finally, the quantization codebook is determined with k-means clustering. Numerical simulations show that our strategy improves the zeroforcing beamforming sum rate by 26.8%, compared with the popular CsiNet. The number of model parameters is reduced by 24.9 times, thus causing a significantly smaller offloading overhead. At the same time, PCA is characterized by a lightweight unsupervised training, requiring eight times fewer training samples than CsiNet.

show abstract

Unsupervised Learning of Adaptive Codebooks for Deep Feedback Encoding in FDD Systems

Cited by 9 publications

References 13 publications

GMM-based Codebook Construction and Feedback Encoding in FDD Systems

GMM-based Codebook Construction and Feedback Encoding in FDD Systems

Centralized Learning of the Distributed Downlink Channel Estimators in FDD Systems using Uplink Data

Machine Learning-Based CSI Feedback With Variable Length in FDD Massive MIMO

Contact Info

Product

Resources

About