Robust Communication and Computation using Deep Learning via Joint Uncertainty Injection

TL;DR

The paper addresses robust joint resource management for a multi-antenna base station serving multiple devices under uncertainties in CSI and computation states. It introduces a DNN-based resource allocator with joint uncertainty injection that maps estimated channel and task requirements to transmit power, computation power, and task allocations, trained to minimize the robust metric $t_ ext{max}^\gamma$ via backpropagation. Key contributions include a gamma-quantile robust objective for joint communication and computation, a softmax-based resource-mapping that respects budget constraints, and an unsupervised training loop that implicitly learns uncertainty distributions, with results showing superior performance of the joint uncertainties approach over schemes that isolate individual uncertainties, especially in high-uncertainty and low-power regimes. This work demonstrates a scalable ML-driven pathway for robust 6G resource management at the network edge, enabling predictable delays in combined communication and computation tasks.

Abstract

The convergence of communication and computation, along with the integration of machine learning and artificial intelligence, stand as key empowering pillars for the sixth-generation of communication systems (6G). This paper considers a network of one base station serving a number of devices simultaneously using spatial multiplexing. The paper then presents an innovative deep learning-based approach to simultaneously manage the transmit and computing powers, alongside computation allocation, amidst uncertainties in both channel and computing states information. More specifically, the paper aims at proposing a robust solution that minimizes the worst-case delay across the served devices subject to computation and power constraints. The paper uses a deep neural network (DNN)-based solution that maps estimated channels and computation requirements to optimized resource allocations. During training, uncertainty samples are injected after the DNN output to jointly account for both communication and computation estimation errors. The DNN is then trained via backpropagation using the robust utility, thus implicitly learning the uncertainty distributions. Our results validate the enhanced robust delay performance of the joint uncertainty injection versus the classical DNN approach, especially in high channel and computational uncertainty regimes.

Figures (6)

• Figure 1: Network setup with multi antenna BS and $4$ devices.
• Figure 2: Proposed DNN architecture with estimated parameters input, resource allocation output, and uncertainty injection.
• Figure 3: Robust worst-case delay $\hat{t}_\text{max}^{\space\gamma}$ as a function of the channel uncertainty, i.e., $\sigma_h^2$.
• Figure 4: Robust worst-case delay $\hat{t}_\text{max}^{\space\gamma}$ as a function of computing uncertainty, i.e., $\sigma_\omega^2$.
• Figure 5: Robust worst-case delay $\hat{t}_\text{max}^{\space\gamma}$ vs. power budget $P^\text{max}$.