Cooperation and Federation in Distributed Radar Point Cloud Processing

TL;DR

The paper quantifies and compares the benefits of radar federation with respect to cooperation mechanisms and proposes a federation mechanism where the radars exchange the parameters of a Bayesian posterior measure of the observed PCs, rather than raw data.

Abstract

The paper considers the problem of human-scale RF sensing utilizing a network of resource-constrained MIMO radars with low range-azimuth resolution. The radars operate in the mmWave band and obtain time-varying 3D point cloud (PC) information that is sensitive to body movements. They also observe the same scene from different views and cooperate while sensing the environment using a sidelink communication channel. Conventional cooperation setups allow the radars to mutually exchange raw PC information to improve ego sensing. The paper proposes a federation mechanism where the radars exchange the parameters of a Bayesian posterior measure of the observed PCs, rather than raw data. The radars act as distributed parameter servers to reconstruct a global posterior (i.e., federated posterior) using Bayesian tools. The paper quantifies and compares the benefits of radar federation with respect to cooperation mechanisms. Both approaches are validated by experiments with a real-time demonstration platform. Federation makes minimal use of the sidelink communication channel (20 ÷ 25 times lower bandwidth use) and is less sensitive to unresolved targets. On the other hand, cooperation reduces the mean absolute target estimation error of about 20%.

Figures (6)

• Figure 1: Model of the proposed federated radar system and its point cloud representation.
• Figure 2: Cooperation v. Federation PC processing architectures: examples from real data. From left to right: (a) scene $\mathbb{S}_{t}$ consist of 2 targets co-present and moving in opposite directions with two radars monitoring the same scene; (b) cooperation mode: the radars exchange pre-processed PCs $\mathbb{\widehat{Q}}_{t,k}$ and obtain a target probability measure via local fusion of the received PCs; (c) federation mode: the radars exchange the parameters $\mathbb{P}_{k}$ characterizing the local posterior measure approximated as Gaussian mixture (number of Gaussian components $M_{k}$, weights $\beta_{k,m}$, mean $\mathbf{\boldsymbol{\mu}}_{k,m}$, and covariance $\boldsymbol{\Gamma}_{k,m}$ of each component $m$).
• Figure 3: Proposed architecture and system model. Radar $k$ monitoring the sidelink (S) and receiving information from radar $h$. The radar MIMO HW supports range-azimuth and elevation tracking and it is equipped with a DSP to extract time-varying 2D/3D point cloud information. The radar CFAR configuration parameters can be controlled via UART communication with the MCU.
• Figure 4: Scene example featuring 2 subjects moving from position L to H and from M to F. Posteriors observed over three consecutive time samples ($t=1,2,3$): a) no-cooperation case with $k=1,2,3$; and b) federation and cooperation cases for $K=3$ radars with deployment described above.
• Figure 5: Kullback-Leibler (KL) $\mathrm{D_{KL}}[\cdot]$ divergence sample probability functions $\mathrm{Pr}[\mathrm{D_{KL}}]$: solid line compares global and federated posteriors $(\mathrm{D_{KL}}[\pi_{t}^{(\mathrm{C})}\Vert\pi_{t}^{(\mathrm{F})}]$), dashed lines compare global and local posteriors for the three deployed radars. Probability measure is obtained by collecting $900$ consecutive sample divergences ($9$ sec. with $10$ms per localization update) corresponding to $2$ subjects moving from position L to H and from M to F (same scenario as in Figure \ref{['scene']}).