Abstract
We propose a novel, learning-based method for adaptively generating lowprobability of detection (LPD) radar waveforms that blend into their operatingenvironment. Our waveforms are designed to follow a distribution that isindistinguishable from the ambient radio frequency (RF) background -- whilestill being effective at ranging and sensing. To do so, we use an unsupervised,adversarial learning framework; our generator network produces waveformsdesigned to confuse a critic network, which is optimized to differentiategenerated waveforms from the background. To ensure our generated waveforms arestill effective for sensing, we introduce and minimize an ambiguityfunction-based loss on the generated waveforms. We evaluate the performance ofour method by comparing the single-pulse detectability of our generatedwaveforms with traditional LPD waveforms using a separately trained detectionneural network. We find that our method can generate LPD waveforms that reducedetectability by up to 90% while simultaneously offering improved ambiguityfunction (sensing) characteristics. Our framework also provides a mechanism totrade-off detectability and sensing performance.