Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Modular Hypernetworks for Scalable and Adaptive Deep MIMO Receivers

Tomer Raviv, Nir Shlezinger

TL;DR

This work tackles the rigidity of static DNN-based deep receivers in dynamic uplink MIMO systems where the number of users $K[t]$ and channel conditions vary. It introduces modular hypernetworks that generate per-user module weights $\{\boldsymbol{\theta}^{(k)}[t]\}$ conditioned on fixed-size embeddings $\boldsymbol{u}^{\text{user}}_k[t]$, enabling the architecture to adapt its complexity as $K[t]$ changes without on-device retraining. Offline training over diverse channel conditions enables rapid on-the-fly adaptation during deployment, with complexity savings compared to online retraining while maintaining near-online performance. The approach offers a scalable, low-latency solution for adaptive deep receivers in mobile networks, balancing accuracy, flexibility, and computational budget.

Abstract

Deep neural networks (DNNs) were shown to facilitate the operation of uplink multiple-input multiple-output (MIMO) receivers, with emerging architectures augmenting modules of classic receiver processing. Current designs consider static DNNs, whose architecture is fixed and weights are pre-trained. This induces a notable challenge, as the resulting MIMO receiver is suitable for a given configuration, i.e., channel distribution and number of users, while in practice these parameters change frequently with network variations and users leaving and joining the network. In this work, we tackle this core challenge of DNN-aided MIMO receivers. We build upon the concept of hypernetworks, augmenting the receiver with a pre-trained deep model whose purpose is to update the weights of the DNN-aided receiver upon instantaneous channel variations. We design our hypernetwork to augment modular deep receivers, leveraging their modularity to have the hypernetwork adapt not only the weights, but also the architecture. Our modular hypernetwork leads to a DNN-aided receiver whose architecture and resulting complexity adapts to the number of users, in addition to channel variations, without retraining. Our numerical studies demonstrate superior error-rate performance of modular hypernetworks in time-varying channels compared to static pre-trained receivers, while providing rapid adaptivity and scalability to network variations.

Modular Hypernetworks for Scalable and Adaptive Deep MIMO Receivers

TL;DR

This work tackles the rigidity of static DNN-based deep receivers in dynamic uplink MIMO systems where the number of users and channel conditions vary. It introduces modular hypernetworks that generate per-user module weights conditioned on fixed-size embeddings , enabling the architecture to adapt its complexity as changes without on-device retraining. Offline training over diverse channel conditions enables rapid on-the-fly adaptation during deployment, with complexity savings compared to online retraining while maintaining near-online performance. The approach offers a scalable, low-latency solution for adaptive deep receivers in mobile networks, balancing accuracy, flexibility, and computational budget.

Abstract

Deep neural networks (DNNs) were shown to facilitate the operation of uplink multiple-input multiple-output (MIMO) receivers, with emerging architectures augmenting modules of classic receiver processing. Current designs consider static DNNs, whose architecture is fixed and weights are pre-trained. This induces a notable challenge, as the resulting MIMO receiver is suitable for a given configuration, i.e., channel distribution and number of users, while in practice these parameters change frequently with network variations and users leaving and joining the network. In this work, we tackle this core challenge of DNN-aided MIMO receivers. We build upon the concept of hypernetworks, augmenting the receiver with a pre-trained deep model whose purpose is to update the weights of the DNN-aided receiver upon instantaneous channel variations. We design our hypernetwork to augment modular deep receivers, leveraging their modularity to have the hypernetwork adapt not only the weights, but also the architecture. Our modular hypernetwork leads to a DNN-aided receiver whose architecture and resulting complexity adapts to the number of users, in addition to channel variations, without retraining. Our numerical studies demonstrate superior error-rate performance of modular hypernetworks in time-varying channels compared to static pre-trained receivers, while providing rapid adaptivity and scalability to network variations.
Paper Structure (17 sections, 2 theorems, 17 equations, 8 figures, 1 table, 1 algorithm)

This paper contains 17 sections, 2 theorems, 17 equations, 8 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. System Model
  3. Communication System
  4. Problem Formulation
  5. Deep Receivers
  6. Hypernetwork Adaptation of Modular Deep Receivers
  7. Modular Hypernetworks
  8. User Embedding
  9. Hypernetwork Adaptation
  10. Hypernetwork Training
  11. Complexity Analysis
  12. Discussion
  13. Numerical Evaluations
  14. Experimental Setup
  15. Weights Only Adaptation
  16. ...and 2 more sections

Key Result

Proposition 1

Consider the transmission of scalar symbols $\mathcal{S}\subset \mathbb{C}$ with pilots holding $B^{\rm pilot} > K^{\max}$. Then, the ratio in the average per-block complexity of online learning and modular hypernetwork adaptation when using a modular architecture with $|{\boldsymbol{\theta}}|$ para

Figures (8)

  • Figure 1: The weights-generation pipeline of modular hypernetworks.
  • Figure 2: Block-varying snr profiles, $K[t]=14$.
  • Figure 3: Synthetic channel, time-invariant $K[t]$.
  • Figure 4: COST 2100 channel, time-invariant $K[t]$.
  • Figure 5: Synthetic channel, time-varying $K[t]$.
  • ...and 3 more figures

Theorems & Definitions (3)

  • Proposition 1
  • proof
  • Corollary 1