Incorporating generated annotated data into training sets can improve object detection. Prior approaches either condition image generation on annotation layouts, limiting diversity and often causing misalignment, or generate images independently and annotate them afterward, reducing accuracy. We introduce a diffusion model that jointly generates images and annotations, enabling their co-evolution and mutual dependency throughout the process. This design achieves tight image-annotation alignment and produces diverse scenarios beyond the original training set, enhancing object detection performance when used in training.

Radar-based object detection is essential for autonomous driving due to radar's long detection range. However, the sparsity of radar point clouds, especially at long range, poses challenges for accurate detection. Existing methods increase point density through temporal aggregation with ego-motion compensation, but this approach introduces scatter from dynamic objects, degrading detection performance. We propose DoppDrive, a novel Doppler-Driven temporal aggregation method that enhances radar point cloud density while minimizing scatter. Points from previous frames are shifted radially according to their dynamic Doppler component to eliminate radial scatter, with each point assigned a unique aggregation duration based on its Doppler and angle to minimize tangential scatter. DoppDrive is a point cloud density enhancement step applied before detection, compatible with any detector, and we demonstrate that it significantly improves object detection performance across various detectors and datasets.

Identifying a pedestrian's intention to cross the road is crucial for autonomous driving, as it alerts the system to stop or slow down. However, determining crossing intention from video is challenging due to the need for extracting complex high-level semantics. This paper introduces ClipCross, a novel classification framework optimized to ex-tract high-level semantic features using the vision-language model CLIP for determining crossing intention. Existing CLIP-based methods perform poorly in this task, as CLIP's image and text encoders fail to capture the nuanced se-mantic distinctions between crossing and non-crossing in-tention images. Clip Cross addresses this by optimizing a set of CLIP text embeddings to extract high-level semantic features, which a multi-layer perceptron uses to distinguish between crossing and non-crossing intentions. Clip Cross achieves state-of-the-art performance on crossing intention estimation benchmark datasets: PIE, PSI, and lAAD.

Object detection in radar imagery with neural networks shows great potential for improving autonomous driving. However, obtaining annotated datasets from real radar images, crucial for training these networks, is challenging, especially in scenarios with long-range detection and ad-verse weather and lighting conditions where radar performance excels. To address this challenge, we present Rad-SimReal, an innovative physical radar simulation capable of generating synthetic radar images with accompanying annotations for various radar types and environmental conditions, all without the need for real data collection. Re-markably, our findings demonstrate that training object de-tection models on RadSimReal data and subsequently eval-uating them on real-world data produce performance lev-els comparable to models trained and tested on real data from the same dataset, and even achieves better performance when testing across different real datasets. Rad-SimReal offers advantages over other physical radar simulations that it does not necessitate knowledge of the radar design details, which are often not disclosed by radar sup-pliers, and has faster run-time. This innovative tool has the potential to advance the development of computer vision al-gorithms for radar-based autonomous driving applications. Our GitHub: https://yuvalhg.github.io/RadSimReal.

The BCJR algorithm is renowned for its optimal equalization, minimizing bit error rate (BER) over intersymbol interference (ISI) channels. However, its complexity grows exponentially with the channel memory, posing a significant computational burden. In contrast, the linear minimum mean square error (LMMSE) equalizer offers a notably simpler solution, albeit with reduced performance compared to the BCJR. Recently, Neural Network (NN) based equalizers have emerged as promising alternatives. Trained to map observations to the original transmitted symbols, these NNs demonstrate performance similar to the BCJR algorithm. However, they often entail a high number of learnable parameters, resulting in complexities comparable to or even larger than the BCJR. This paper explores the potential of NN-based equalization with a reduced number of learnable parameters and low complexity. We introduce a NN equalizer with complexity comparable to LMMSE, surpassing LMMSE performance and achieving a modest performance gap from the BCJR equalizer. A significant challenge with NNs featuring a limited parameter count is their susceptibility to converging to local minima, leading to suboptimal performance. To address this challenge, we propose a novel NN equalizer architecture with a unique initialization approach based on LMMSE. This innovative method effectively overcomes optimization challenges and enhances LMMSE performance, applicable both with and without turbo decoding.

Automotive radars have an important role in autonomous driving systems. The main challenge in automotive radar detection is the radar’s wide point spread function (PSF) in the angular domain that causes blurriness and clutter in the radar image. Numerous studies suggest employing an ’end-to-end’ learning strategy using a Deep Neural Network (DNN) to directly detect objects from radar images. This approach implicitly addresses the PSF’s impact on objects of interest. In this paper, we propose an alternative approach, which we term ’’Boosting Radar Reflections" (BoostRad). In BoostRad, a first DNN is trained to narrow the PSF for all the reflection points in the scene. The output of the first DNN is a boosted reflection image with higher resolution and reduced clutter, resulting in a sharper and cleaner image. Subsequently, a second DNN is employed to detect objects within the boosted reflection image. We develop a novel method for training the boosting DNN that incorporates domain knowledge of radar’s PSF characteristics. BoostRad’s performance is evaluated using the RADDet and CARRADA datasets, revealing its superiority over reference methods.

Estimating the directions of arrival (DOAs) of multiple sources from a single snapshot obtained by a coherent antenna array is a well-known problem, which can be addressed by sparse signal reconstruction methods, where the DOAs are estimated from the peaks of the recovered high-dimensional signal. In this paper, we consider a more challenging DOA estimation task where the array is composed of non-coherent sub-arrays (i.e., sub-arrays that observe different unknown phase shifts due to using low-cost unsynchronized local oscillators). We formulate this problem as the reconstruction of a joint sparse and low-rank matrix and solve its convex relaxation. While the DOAs can be estimated from the solution of the convex problem, we further show how an improvement is obtained if, instead, one estimates from this solution only the phase shifts, creates "phase-corrected" observations and applies another final (plain, coherent) sparsity-based DOA estimation. Numerical experiments show that the proposed approach outperforms strategies that are based on non-coherent processing of the sub-arrays as well as other sparsity-based methods.

In this paper we develop a deep neural network (DNN) method for estimating the object surface from radar 2D image (azimuth-range). The DNN is designed to maintain the input image angular resolution and produces two outputs per each angle, which are a classification bit and a regression value. The classification bit determines whether there is a reflection point per each angle, and the regression value is the estimated reflection range. We have developed a statistical simulation model that approximates the statistics of the radar image, and trained the network with synthetically generated examples from the simulation model. The network showed good performance on radar images that were obtained from real radar measurements, and also showed to outperform the common CFAR reference method.

We consider the problem of estimating the angles of arrival (AOAs) of multiple sources from a single snapshot obtained by a set of non-coherent sub-arrays, i.e., while the antenna elements in each sub-array are coherent, each sub-array observes a different unknown phase. Previous relevant works are based on eigendecomposition of the sample covariance, which requires a large number of snapshots, or on combining the sub-arrays using non-coherent processing methods. In this paper, we propose a technique to estimate the sub-arrays phase offsets for a given AOAs hypothesis, which facilitates approximate maximum likelihood estimation of the AOAs from a single snapshot. Numerical experiments show that the proposed approach clearly outperforms non-coherent processing, and even attains the Cramér-Rao lower bound in various scenarios.

The problem of estimating the number of sources and their angles of arrival from a single antenna array observation has been an active area of research in the signal processing community for the last few decades. When the number of sources is large, the maximum likelihood estimator is intractable due to its very high complexity, and therefore alternative signal processing methods have been developed with some performance loss. In this paper, we apply a deep neural network (DNN) approach to the problem and analyze its advantages with respect to signal processing algorithms. We show that an appropriate designed network can attain the maximum likelihood performance with feasible complexity and outperform other feasible signal processing estimation methods over various signal to noise ratios and array response inaccuracies.

In this paper a multi-radar beamforming algorithm is developed that attains higher angular resolution than the individual radars. The method is based on jointly processing the received signals from mutually non-coherent radars that are widely spaced on a vehicle. The proposed method can separate and discriminate between close targets, which are not separated by trilateration nor by filtering the multiple radars measurements.

Attaining accurate estimation of a signal time-of-arrival (TOA) in dense multipath channels is very challenging. This problem was traditionally solved with signal processing techniques. In this paper, a novel deep convolutional neural networks (DCNN) TOA estimator is developed. The DCNN was trained with synthetically generates multipath channel realizations based on statistical modeling, and then tested on real-life measurements from indoor environments. It is shown that the DCNN attains the performance of state-of-the-art signal processing based estimators (maximum likelihood performance), and has the advantage that it does not require knowledge of the channel statistics nor the knowledge of the transmitted signal waveform. This work inspires further study on the applicability of neural networks to other related problems, which have been traditionally solved with signal processing methods.

This paper presents a novel analytical expression which approximates the bit error rate (BER) of the joint phase and symbol maximum a posteriori (JMAP) sequence estimator for: trellis code modulation (TCM) with M-ary phase shift keying (MPSK) modulation; any arbitrary phase noise model (i.e., not limited to the Wiener process); and either a matched or mismatched decoder. First, we derived convenient closed-form expressions for approximating the pairwise error probability of two code sequences for both matched and mismatched decoders. Since the expressions are formulated either in the time or in the frequency domain, it is possible to indicate the contribution of every frequency in the phase noise spectrum. We then applied the union bound on the code sequence pairwise errors. The analytical expression was tight (usually <;1dB) for MPSK constellations with M ≥ 4 and code rates ≥ 0.5. Once developed, the expressions will assist designers to consider the influence on receiver performance of the code characteristics, decoder implementation and RF synthesizer phase noise. It further enables joint optimization of the RF synthesizer, the code and the decoder for achieving the lowest error rate or other design targets.

As communication extends to higher carrier frequencies, the phase noise problem becomes more severe and conventional phase tracking methods become inadequate. Jointly maximum likelihood sequence estimation (JMLSE) phase tracking and decoding is a practical tracking method achieving near optimal performance. There is a lack of an analytical tool for performance evaluation of coded JMLSE. In this paper we will present a novel approximated union bound on the performance of the JMLSE receiver for trellis coded MPSK and Wiener phase noise. Our analysis leads to an equivalent model of the JMLSE process which gives important insights. Since the union bound requires the summation of infinite number of error event, we introduce an efficient algorithm for selecting the error events with the significant contribution. The developed tool is usually tight and efficient, hence can be instrumental in finding codes which achieve low error rate on high phase noise channels.