The lensless camera, which replaces the camera’s lens with a thin coded mask, can be manufactured to be lightweight, compact, and less expensive compared to traditional lens-based cameras. Furthermore, when employing the radial mask proposed by J. Neto et al., it becomes possible to capture very deep depth-of-field images without narrowing the aperture, a feat challenging to achieve with conventional lens imaging [Paper ID: 10439, Title: Extended Depth-of-Field Lensless Imaging Using an Optimized Radial Mask]. Previously, this imaging method was tested with an industrial board-level camera, and the basic experiments were conducted inside a darkroom. In this demonstration, we implement this technology in a mirrorless digital camera and show its utility through photography outside of the darkroom. Additionally, while coded masks using Spatial Light Modulator (SLM) have been used so far, this demo uses mask made using chrome-deposited glass which is a brighter, thinner, and higher-resolution coded optics than the SLM.

Analog-to-digital converters (ADCs) are vital in digital signal processing, but their cost and energy consumption rise with increased sampling rates, making it advantageous to sample at the lowest effective rate. The dynamic range of ADCs is also crucial, ideally surpassing the input signal's range to avoid clipping and data loss. However, a larger dynamic range can lead to increased quantization error, affecting accuracy. To address this, one can apply a modulo operation to the input signal before sampling if it exceeds the ADC's dynamic range. Modulo sampling, particularly for band-limited (BL) signals, has been thoroughly studied [1,2]. Realistic ADCs often have built-in low-pass filters (LPFs) with cutoff frequencies linked to their maximum sampling rate. While this setup usually filters out noise, it poses challenges for signals like multi-band ones, where crucial information may lie in the filtered-out frequency range. In modulo sampling, a BL input signal could broaden its bandwidth or lose its BL characteristics post-modulation. Most existing research assumes an ideal sampler with the capability to accurately measure amplitude in a wide band. Practically, this necessitates an ADC with a much higher sampling capacity than what is typically used, thus treating the signal as BL with a vast bandwidth. Our demonstration presents the first hardware capable of modulo sampling with a realistic sampler, complete with the recovery process. We exhibit a live modulo operation on a BL signal using three methods: an ideal sampler with established recovery techniques, the challenges with a realistic sampler, and our proposed system with a realistic sampler. Our goal is to show live modulo recovery using a practical ADC, moving away from the reliance on ideal samplers that need far greater capabilities than necessary for the task. Our approach with a realistic sampler achieves this without requiring such excessive capability.

Detailed description of the demo: This work will showcase a real-time demonstration of high dynamic range FMCW radar that is based on the Unlimited Sensing Framework. The radar baseband signals will be sampled using the US-ADC, and the reconstruction and processing of the data will be performed live. The hardware presented will be an FMCW radar, unlimited sampling ADCs that are interfaced with another acquisition board, and a computer that will show both the classic acquisition (and its possible limitations) and the unlimited samples and the live reconstruction of the radar signal. Main novelty and innovations of the demo: This live demo will be the first time Unlimited Sampling is presented as a practical solution for a real-time application on a commercial sensor, tackling multiple signals' simultaneous acquisition and reconstruction. Impact to signal processing communities: Unlimited sampling theory has the potential to change how acquisition is performed and alleviate some limitations commonly found, such as saturations/clipping. Consequently, this practical live demo will highlight the potential that Unlimited Sampling has for the signal processing community. More specifically, this demo will show that HDR reconstruction from folded radar measurements using multiple channels can be achieved at a lower sampling rate than other standard methods. Interactivity for both on-site and online attendees: On-site & Online: the demo will be performed live, and the attendees will be able to interact with the radar front-end and see the effect on the Unlimited Samples and the reconstruction similar to this live demo (https://www.youtube.com/watch?v=cENWT5mQDXA&t=563s). Moving a corner reflector in front of the radar will change the signal’s frequency content of the sampled signals, and the attendees will be able to see the reconstruction performed live.

Gold nanoparticles (GNPs) hold promise to improve the detection and treatment of diseases such as cancer and Alzheimer’s. However, detecting GNPs in the body remains an unmet technological challenge. This demonstration is based on our work introducing a new methodology for remotely localizing and tracking GNPs in various therapeutic settings, including through biological tissues, using a compact millimeter-wave (mmWave) radar system [1]. By modeling GNP detection as a sparse recovery problem, we propose a method that localizes GNPs in cluttered environments as well as detects changes in concentrations that relate to changes in the intensity of the reflected signals. Our approach paves the way to a first-ever noninvasive, non-ionizing, simple radar-based approach for bioimaging and tracking GNPs inside the body. In this demonstration, we propose a dedicated platform in which a mmWave radar is placed on a table a few centimeters away from a brain model containing a solution with a known concentration of GNPs. The radar transmits a series of chirp signals towards the GNP concentration. Then, by processing the reflected echo signals, we examine the effect of varying GNP concentrations on the measured distance, angle and amplitude, from the reflecting concentration. Our demonstration platform is based on a dedicated experimental setup of an FMCW mmWave radar sensor (TI IWR1443 77-81 GHz) and a brain model based on agarose gel in a shape of a brain with a glass vile inserted into a specific spot. This allows to test the response to different GNP concentrations through a structure that mimics brain tissue. By using a dedicated GUI, both on-site and online attendees can evident the localization and tracking of GNPs using mmWave FMCW radar utilizing our proprietary algorithm. [1] Y. Eder, R. Abel, A. Schroeder, and Y.C. Eldar, "Localization and Tracking of Gold Nanoparticles Using mmWave FMCW Radar”, to appear in ICASSP 2024.

Although drones have been widely used for several applications, their exponential rise has caused security and privacy concerns in the recent past. For critical applications such as defense and airline operations, there is a demanding need to develop real-time drone detection and tracking systems for identifying intruding drones and neutralizing their harmful effects using appropriate countermeasures. Current approaches lack real-time performance, have a high false negative rate, and are not generalizable across different imaging modalities. To mitigate these limitations, we propose IR-RGB SETNET which is a fast, robust, and generalizable network for drone detection and tracking deployable on resource-constrained platforms. IR-RGB SETNET is an ensemble of base networks from the YOLOv5 object detection family that are compressed using a combination of network pruning, quantization, and knowledge distillation and are trained on infrared and RGB video sequences. The network also has a tracking module with the Deep-SORT algorithm at its core. Through this demonstration, we showcase IR-RGB SETNET’s superior 5-fold improvement in inference speed and its robustness in drone localization under severe distortions, lighting variations, and domain randomizations. Our demo consists of a hardware setup involving an NVIDIA Jetson Nano interfaced with a network of cameras. We plan to deploy IR-RGB SETNET on the edge and showcase real-time experiments to gauge the effect of different compression algorithms on the network’s performance. We also envision showcasing a real-time surveillance system that sends critical alerts to the control room when a malicious drone is spotted. To achieve this, we plan to add an anomaly detection module to IR-RGB SETNET that involves a vision transformer trained to identify harmful payloads onboard. The idea for this demonstration was inspired by our previous work presented at ICASSP 2023 Workshops and ICIP 2023 Grand Challenge Session.

Unlimited sampling (US) is a computational sensing paradigm for high-dynamic range (HDR) acquisition of signals. In our accepted ICASSP 2024 paper entitled “Neuromorphic sensing meets unlimited sampling,” (Paper ID: 7762) we propose a new technique for US using a neuromorphic encoder (NE). An NE is an opportunistic device that records signals as a stream of events denoting a change in the input by a constant called the contrast threshold, and the signal is represented as a sequence of 2-tuples, each constituting the time instant and the polarity of change. In our proposed neuromorphic US (NUS) scheme, we use an NE to fold the signal that lies outside the dynamic range and simultaneously record a compressed representation of the error signal. This allows the analog-to-digital converter (ADC) to operate at the Nyquist rate of the signal. As compared to modulo sampling, our NUS scheme does not require oversampling. Additional measurements are made using the NE opportunistically, i.e., only when the signal goes beyond the dynamic range while consuming much less power than the uniform sampling ADC. We also show that the maximal rate of additional measurements that may be obtained decays with the dynamic range. Complementary to US, we provide a perfect reconstruction strategy that only requires addition of a residual signal constructed using the compressed representation of the error signal. We show that perfect reconstruction can be achieved in real time. Our proposal for the show-and-tell demo comprises a hardware demonstration of NUS. Our NE is designed in-house and involves simple and cost-effective components. NUS is such that the NE is plugged into an existing ADC to achieve US and does not require specialised hardware such as the self-reset ADC. In hardware, we demonstrate HDR acquisition and real-time reconstruction of synthetic signals. We show that NUS operates at the Nyquist rate of the signal, while opportunistically recording compressed measurements.

In high-dynamic range (HDR) analog-to-digital converters (ADCs), quantization bits are highly sought to keep the quantization error low. However, this leads to an elevated bit rate, undesirable in numerous applications. To address this issue, a strategy combining modulo-folding with a low-DR ADC can create an efficient HDR-ADC with fewer bits. However, this approach typically requires oversampling and a high bit rate. Further, modulo-folding hardware typically requires multi-level power sources whose maximum value is comparable to the HDR input signal. This makes the modulo-folding-based HDR-ADC less attractive for devices with a limited power budget. Recently, we introduced an alternative approach for achieving HDR-ADC based on phase modulation. In this method, we employ the analog signal committed for sampling to modulate the phase of a sinusoidal carrier signal. This modulation effectively constrains the DR of the resulting signal, allowing for the utilization of a low-DR ADC with fewer bits. A few prominent features of the proposed approach are: 1. The sampling rate could be minimal and does not require oversampling, especially in the absence of noise. 2. In the presence of noise, we demonstrate that the proposed PM-based HDR approach exhibits efficiency, characterized by lower reconstruction errors and reduced sampling rates compared to modulo-based HDR-ADCs. Regarding hardware realization, the HDR-ADC requires a single-domain power supply whose value is the same as the low-DR-ADC’s dynamic range, smaller than the analog signal’s DR. In this demonstration, we present our hardware prototype, showing the proposed framework's reconstruction ability for different input signal amplitude levels. Further, through hardware realization, we compare the power efficiency of a modulo-based HDR-ADC and the proposed modulation-based HDR-ADC. This novel framework can replace existing high-bit rate HDR-ADCs while maintaining the existing bit rate requirements.

The Unlimited Sensing Framework (USF) [1,2] is an emerging digital technology that offers advantages over conventional analog-to-digital converters (ADCs). During ICASSP 2021, we introduced the inaugural implementation of the modulo ADC within the USF, illustrating its capability to capture high dynamic range (HDR) signals. In the subsequent ICASSP 2022, we explored the integration of modulo ADCs with radars, unveiling novel applications. Building on this progress, ICASSP 2023 featured the introduction of the Multi-Channel (MC) modulo ADC system, highlighting the potential of USF-multichannel channel systems to novel Signal Processing applications. Sub-Nyquist spectral estimation of high bandwidth signals is a well-studied topic. The approaches developed struggle with high-dynamic-range (HDR) signals that can saturate analog-to-digital converters (ADCs), yielding acquisition bottlenecks. To overcome this barrier, in this demo, we demonstrate a sub-Nyquist spectral estimation system based on the novel Unlimited Sensing Framework (USF). Our method is theoretically guaranteed, resulting in an exact K-frequency estimation from only 6K+4 measurements, independent of the sampling rate or folding threshold. Utilizing multi-channel modulo-ADCs, in realistic scenarios, we show that our method enables successful spectrum estimation of HDR signals in the kilohertz range at a sampling rate of hertz scale (1.45% Nyquist rate). In this demo, we show a 33-fold improvement in frequency estimation accuracy using one less bit compared to the conventional ADCs. We show an end-to-end implementation that entails on-the-fly signal acquisition and recovery, where system parameters such as frequency of the input signal, sampling rate, and folding threshold, are tunable for the audience. [1] Bhandari et al. On Unlimited Sampling and Reconstruction (IEEE TSP, 2020) [2] Bhandari et al. Unlimited Sampling from Theory to Practice (IEEE TSP, 2021)

The Unlimited Sensing Framework (USF) [1,2] is an emerging digital technology that offers advantages over conventional analog-to-digital converters (ADCs). During ICASSP 2021, we introduced the inaugural implementation of the modulo ADC within the USF, illustrating its capability to capture high dynamic range (HDR) signals. In the subsequent ICASSP 2022, we explored the integration of modulo ADCs with radars, unveiling novel applications. Building on this progress, ICASSP 2023 featured the introduction of the Multi-Channel (MC) modulo ADC system, highlighting the potential of USF-multichannel channel systems to novel Signal Processing applications. Capturing extreme HDR signals with high bandwidth has historically been challenging and costly. Our demo in ICASSP 2021 demonstrated the HDR capability of capturing 15x the ADC’s maximum voltage in the sub-kilohertz range. In this demo, our newly designed system can now achieve over 30x folding of HDR signals up to 20-kilohertz bandwidth, illustrating the rapid evolution and scalability of USF hardware. In the live demo, we provide an interactive system that enables a tangible implementation of USF for real-world applications, such as radar communications and sensor arrays. Complemented with the hardware, a robust recovery algorithm is presented to unfold the modulo measurements, exhibiting the full pipeline of acquisition and reconstruction in a real-time fashion. Overall, the demo hopes to give a true flavor of live HDR acquisition, providing a more intuitive sense of modulo sampling. [1] Bhandari et al. On Unlimited Sampling and Reconstruction (IEEE TSP, 2020) [2] Bhandari et al. Unlimited Sampling from Theory to Practice (IEEE TSP, 2021)

Robotic swarm with portable sensor nodes is an emerging technology for sensing dynamic physical processes both on Earth and in future space exploration missions. These nodes are equipped with radio transceivers, providing precise time and position references without additional infrastructures like GPS. Each node is additionally equipped with environmental sensors, for example a photonic sensor to sense the illumination in caves beneath the lunar surface, a hydrogen sulfide sensor to explore the volcanic activities, or a methane sensor to track organic traces. At the German Aerospace Center (DLR), we design compact and robust sensor “eggs” with ultra-wideband (UWB) technology for decentralized position and environmental estimation. These eggs can be easily moved and dropped by robots to favorable positions, and are thus suitable for technology demonstrations in space-analog missions, for example on the Vulcano Island, Italy, for volcanic gas source sensing, and in a lava cave on Lanzarote, Spain, for cave structure sensing. These eggs are also served as educational platforms, assist students to gain intuitions and in-depth knowledge in signal processing for communications, localization, decentralized estimation, sensor fusion, information gathering, path planning, etc. We propose a live Show and Tell Demo with our sensor eggs equipped with photonic sensors. Eggs’ positions and illumination field are estimated and visualized in real-time. Audiences are welcome to interact with the eggs, e.g., changing the lighting condition, moving the eggs or even altering the self-organized network or the decentralized estimators. In the meantime, we present slides or posters to provide supplementary information on our fundamental research and space-analog missions at DLR. We believe it will be a fruitful event with exciting entertainment, stimulating discussions and a glimpse on many signal processing aspects in future autonomous space exploration missions.