Channel-Aware Throughput Maximization for Cooperative Data Fusion in CAV

TL;DR

The paper tackles cooperative perception in connected and autonomous vehicles under dynamic wireless channels that constrain data fusion throughput. It introduces TMAC, a channel-aware throughput maximization framework that pairs a self-supervised autoencoder for adaptive data compression with a two-stage MIP decomposition (P1–P2) to optimize data rates, compression, and link establishment, validated on the OpenCOOD/OPV2V platform. Key contributions include a linearized subproblem (P1-2) solution for rate and compression, a bi-level optimization for link selection, a fine-tuning strategy to leverage historical data and reduce spectral resources by 42%, and substantial improvements in both network throughput ($>20 ext{\%}$) and perception accuracy (AP@IoU $>9 ext{\%}$) with latency under 100 ms. The approach offers a practical pathway to reliable, high-throughput cooperative perception in real-world V2V networks while accommodating time-varying channels and bandwidth constraints.

Abstract

Connected and autonomous vehicles (CAVs) have garnered significant attention due to their extended perception range and enhanced sensing coverage. To address challenges such as blind spots and obstructions, CAVs employ vehicle-to-vehicle (V2V) communications to aggregate sensory data from surrounding vehicles. However, cooperative perception is often constrained by the limitations of achievable network throughput and channel quality. In this paper, we propose a channel-aware throughput maximization approach to facilitate CAV data fusion, leveraging a self-supervised autoencoder for adaptive data compression. We formulate the problem as a mixed integer programming (MIP) model, which we decompose into two sub-problems to derive optimal data rate and compression ratio solutions under given link conditions. An autoencoder is then trained to minimize bitrate with the determined compression ratio, and a fine-tuning strategy is employed to further reduce spectrum resource consumption. Experimental evaluation on the OpenCOOD platform demonstrates the effectiveness of our proposed algorithm, showing more than 20.19\% improvement in network throughput and a 9.38\% increase in average precision (AP@IoU) compared to state-of-the-art methods, with an optimal latency of 19.99 ms.

Figures (6)

• Figure 1: An example of a vehicle-to-vehicle (V2V) network consisting of four connected autonomous vehicles (CAVs). Fig. \ref{['fig:system model']}(a): CAV0 is the ego vehicle that can incorporate the viewpoints of CAV1 and CAV3. However, the link between CAV0 and CAV2 is disconnected to avoid negative impact on the overall throughput. Fig. \ref{['fig:system model']}(b): Each CAV collects the traffic status by four cameras. Fig. \ref{['fig:system model']}(c): Nearby CAVs encode camera data and then transmit them to the Ego CAV for decoding images. Fig. \ref{['fig:system model']}(d): The ego CAV predicts the bird eye's view (BEV) by fusing reconstruction data.
• Figure 2: The overall architecture: 1) $\beta^{*}$ is obtained using Algorithm \ref{['alg:tmac-throughput']} based on the current channel conditions. 2) CAV1 and CAV2 fine-tune a small portion of historical images by updating parameters from roadside units. 3) CAVs use their encoders to convert images into a bitstream, which is then transmitted to the ego CAV. 4) The ego CAV can decode the received bitstream to reconstruct the images, while the reconstructed images are fused together in a Fusion Net to obtain BEV prediction.
• Figure 3: Results in average network throughput under different communication parameters.
• Figure 4: Latency of TMAC when transmitting a data packet in four different settings. $i/10$ denotes that the first $i$ ($i=1$ and $2$) frames out of ten is utilized for fine-tuning, with the remaining frames being compressed before being transmitted to the ego CAV. The term 'best' represents the most favorable outcome achieved in our experiments, while 'worst' refers to the least desirable result.
• Figure 5: Comparison of bandwidth-saving performance between finetuned TMAC and unfinetuned TMAC at (a) the same MS-SSIM level, (b) the same PSNR level.