Federated Dropout Learning for Hybrid Beamforming with Spatial Path Index Modulation in Multi-User Mmwave-Mimo Systems

Abstract

Millimeter wave multiple-input multiple-output (mmWave-MIMO) systems with small number of radio-frequency (RF) chains have limited multiplexing gain. Spatial path index modulation (SPIM) is helpful in improving this gain by utilizing additional signal bits modulated by the indices of spatial paths. In this paper, we introduce model-based and model-free frameworks for beamformer design in multi-user SPIM-MIMO systems. We first design the beamformers via model-based manifold optimization algorithm. Then, we leverage federated learning (FL) with dropout learning (DL) to train a learning model on the local dataset of users, who estimate the beamformers by feeding the model with their channel data. The DL randomly selects different set of model parameters during training, thereby further reducing the transmission overhead compared to conventional FL. Numerical experiments show that the proposed framework exhibits higher spectral efficiency than the state-of-the-art SPIM-MIMO methods and mmWave-MIMO, which relies on the strongest propagation path. Furthermore, the proposed FL approach provides at least 10 times lower transmission overhead than the centralized learning techniques.

Federated Learning for Physical Layer Design

“…Thus, the use of DL is significant for communication-efficiency. It is reported in [13] that FL with (without) DL enjoys approximately 10 (5) times lower overhead than CL.…”

“…Each user possesses the trained model and can estimate its own channel with it Data collection requires a labeling process to obtain the channel data Beamforming in Massive MIMO [5] Gradient transmission for a classification CNN model Simple beamformer construction from the predicted indices at the classification layer Sub-optimum performance due to codebook-based beamformer design Beamforming With SPIM [13] Gradient transmission for a regression CNN model with dropout layers Higher spectral efficiency can leverage the performance loss due to distributed training Labeling requires heavy computation resources Beamforming in IRS-aided Massive MIMO [14] Model transmission for a regression MLP model…”

“…To improve spectral efficiency in mm-Wave systems, index modulation (IM) is an attractive technique, in which additional information is encoded in the indices of the transmission media such as, subcarriers, antennas or spatial paths. FLbased beamforming with spatial path index modulation (SPIM) is considered in [13], where a regression CNN model with dropout layers (DLs) is trained. DLs are useful to mitigate overfitting and for the generalization of the learning model.…”

“…The major drawback of [13] is the labeling stage, which is conducted via alternating manifold optimization (MO) techniques, which may require heavy computational resources, especially for a high number of antennas and RF chains, which is the case in massive MIMO systems. A codebook-based approach can be followed to reduce the cost of labeling while sacrificing from the spectral efficiency [5].…”

“…While an overview of FL has been provided in the relevant literature, e.g., [6,11,12], they either consider nonphysical layer applications, such as image/speech/text recognition/classification [6], or a general FL framework for wireless communications [11,12]. In contrast, this work focuses on the physical layer design and provides a synopsis of various FL-based applications, such as symbol detection [9], channel estimation [10], and beamforming [5,13,14]. In the remainder of this article, we first briefly discuss how FL works and what distinguishes it from CL along with its advantages and drawbacks.…”