This work characterises the effect of mutual interference in a planar network of pulsed-radar devices. Using stochastic geometry tools and a strongest interferer approximation, we derive simple closed-form expressions that pinpoint the role played by key system parameters on radar detection range and false alarm rate in the interferencelimited region. The fundamental tradeoffs of the system between radar performance, network density and antenna directivity are captured for different path-loss exponents in the no-fading and Rayleigh-fading cases. The discussion highlights practical design hints for tuning the radar parameters. The accuracy of the model is verified through network simulations, and the role of random noise on detection in sparse, non interference-limited networks is characterised.
Index TermsRadar, Stochastic Geometry, Interference
I. INTRODUCTIONCompact low-cost radar devices are set to become pervasive for providing short-range environmental awareness in emerging applications such as automotive [1], enhanced localization [2], or radio resource optimization [3]. In future heterogeneous networks, low-power radars are also envisioned to share spectrum -and possibly be co-located -with communication devices [2], [4], in e.g. the 60 GHz unlicensed band. These scenarios give rise to coexistence of a multitude of radar devices, randomly oriented over a large area, sharing a frequency band in an uncoordinated fashion. For such radar networks, it is paramount to understand the effect of mutual interference on the achievable detection and false alarm rates.Mutual radar interference has been thoroughly studied in simple two-node topologies [5]. Recent research has started to address more general configurations. The first results were reported in [6], focusing on OFDM radars. The achievable detection probability for different network densities was investigated, relying on a Gaussian approximation of the aggregate interference. A step forward was taken in [1], with the introduction of a stochastic geometry framework to study the performance of a linear automotive radar network. The authors considered an SIR-based detection model, and derived results under different statistical distributions for the radar positions, only for the A. Munari and L. Simić are with the