Periodic Reporting for period 5 - BNYQ (Breaking the Nyquist Barrier: A New Paradigm in Data Conversion and Transmission)
Berichtszeitraum: 2020-09-01 bis 2021-06-30
Digital signal processing (DSP) is a revolutionary paradigm shift enabling processing of physical data in the digital domain. However, state-of-the-art analog-to-digital convertors (ADCs) preclude high-rate wideband sampling and processing with low cost and energy consumption, presenting a major bottleneck. Our ambitious goal is to introduce a paradigm shift in ADC design that will enable systems capable of low-rate, wideband sensing and low-rate DSP. We challenge current practice that separates the sampling stage from the processing stage and exploit structure in analog signals already in the ADC, to drastically reduce the sampling and processing rates. Our results show that this allows substantial savings in sampling and processing rates: For example, we show rate reduction of 1/28 in ultrasound imaging, and 1/30 in radar detection.
To achieve our overreaching goal we focused on three interconnected objectives developing the 1) theory 2) hardware and 3) applications of sub-Nyquist sampling. Our methodology ties together two areas on the frontier of signal processing: compressed sensing (CS), focused on finite length vectors, and analog sampling. Our research also inherently relies on advances in several other important areas within signal processing and combines multi-disciplinary research at the intersection of signal processing, information theory, optimization, estimation theory and hardware design.
To achieve our overreaching goal we focused on three interconnected objectives developing the 1) theory 2) hardware and 3) applications of sub-Nyquist sampling. Our methodology ties together two areas on the frontier of signal processing: compressed sensing (CS), focused on finite length vectors, and analog sampling. Our research also inherently relies on advances in several other important areas within signal processing and combines multi-disciplinary research at the intersection of signal processing, information theory, optimization, estimation theory and hardware design.
In our research project we have been developing new methods for super-resolution imaging, in both fluorescence microscopy and contrast enhanced ultrasound imaging. This dramatic improvement in the temporal resolution paves the way to both sub-diffraction live-cell imaging in microscopy and clinical sub-diffraction diagnosis in ultrasound. Another area we are working on is low-rate acquisition and processing framework based on frequency domain processing for 3D ultrasound imaging. The method allows obtaining optimal image quality with improved SNR and higher penetration depth, while keeping computational load low enough to enable real time implementation. We also considered a generalization of low-rate frequency domain processing for ultrafast ultrasound imaging based on coherent plane-wave compounding. Another aspect we are working on in the area of ultrasound is improving quality and performance of medical ultrasound systems using a small number of antenna elements.
Moving on to more basic sampling paradigms, we have explored-Sub-Nyquist sampling of correlated signals: A variety of interesting theories have been developed for sampling and reconstruction of individual signals with certain structure but very few results have been reported on sampling a set of signals with a certain underlying correlation. In the area of radar we have considered various methods to achieve high resolution radar using a small number of resources in time, space and frequency.
In the interface of information theory and sampling theory we have recently begun exploring Task-based quantization. Massive multiple-input multiple-output (MIMO) systems have been drawing considerable. We quantify the gain in spectral efficiency of joint-decoding compared to the standard approach of separate linear decoding, and show that the standard approach fails to capture the actual achievable rates of massive MIMO systems. In our work we address the challenges of hardware complexity and pilot contamination.
One of the research areas is Deep Cognitive Sparse Arrays for Automotive Radars. In this case the high-resolution direction of arrival (DOA) estimation requires a large number of antenna elements. In order to balance between hardware complexity and resolution, we proposed a cognitive, scalable, sparse array selection technique based deep neural networks. In this demo, we present a design and implementation of a hardware prototype that demonstrate the proposed sparse antenna selection strategy. Through real-time experiments, we show that the proposed sparse selection method results in a 2-3 dB lower error compared to a typically employed random selection method.
In the area of autonomous vehicles, we came with a Joint Radar-Communications System demo. Our solution implements both radar and communications within the same domain and the same resources. Our alternative strategy, which is the focus of growing research attention, is to handle them as a dual function radar-communications (DFRC) system. Such joint designs improve performance by facilitating coexistence, as well as contributes to reducing the number of antennas, system size, weight and power consumption.
In the area of radars, we came with the Sub-Nyquist Radar with Distorted Pulse Shape demo. Sub-Nyquist systems uses the knowledge of the transmit pulse and the receive signal model to estimate the targets from low-rate samples. In this demo, we build a hardware prototype to demonstrate the proposed sub-Nyquist radar. We show that while operating at 10 times below the Nyquist rate the proposed two-receiver system with unknown pulse has comparable performance to a single-receiver sub-Nyquist system with a known pulse.
Moving on to more basic sampling paradigms, we have explored-Sub-Nyquist sampling of correlated signals: A variety of interesting theories have been developed for sampling and reconstruction of individual signals with certain structure but very few results have been reported on sampling a set of signals with a certain underlying correlation. In the area of radar we have considered various methods to achieve high resolution radar using a small number of resources in time, space and frequency.
In the interface of information theory and sampling theory we have recently begun exploring Task-based quantization. Massive multiple-input multiple-output (MIMO) systems have been drawing considerable. We quantify the gain in spectral efficiency of joint-decoding compared to the standard approach of separate linear decoding, and show that the standard approach fails to capture the actual achievable rates of massive MIMO systems. In our work we address the challenges of hardware complexity and pilot contamination.
One of the research areas is Deep Cognitive Sparse Arrays for Automotive Radars. In this case the high-resolution direction of arrival (DOA) estimation requires a large number of antenna elements. In order to balance between hardware complexity and resolution, we proposed a cognitive, scalable, sparse array selection technique based deep neural networks. In this demo, we present a design and implementation of a hardware prototype that demonstrate the proposed sparse antenna selection strategy. Through real-time experiments, we show that the proposed sparse selection method results in a 2-3 dB lower error compared to a typically employed random selection method.
In the area of autonomous vehicles, we came with a Joint Radar-Communications System demo. Our solution implements both radar and communications within the same domain and the same resources. Our alternative strategy, which is the focus of growing research attention, is to handle them as a dual function radar-communications (DFRC) system. Such joint designs improve performance by facilitating coexistence, as well as contributes to reducing the number of antennas, system size, weight and power consumption.
In the area of radars, we came with the Sub-Nyquist Radar with Distorted Pulse Shape demo. Sub-Nyquist systems uses the knowledge of the transmit pulse and the receive signal model to estimate the targets from low-rate samples. In this demo, we build a hardware prototype to demonstrate the proposed sub-Nyquist radar. We show that while operating at 10 times below the Nyquist rate the proposed two-receiver system with unknown pulse has comparable performance to a single-receiver sub-Nyquist system with a known pulse.
As outlined above, we have made progress on many of the goals of this project, developing a wide array of systems. Improving power, size, speed, cost, data volume, and performance.
The outcome of the research has been implemented in hardware and software demo systems, showing the advantages over traditional approach. These demos have been presented world wide in a large series of conferences.
Areas of interest:
Radar
Communication
Ultrasound
Microscopy
Our results include more efficient and refined algorithms for the above list, based on Deep Learning which we have begun intergrating into our methods. This will provide more effective and faster algorithms we stated above. We are also improving our hardware implementations and integrating them to larger system-wide operations that include several developments combined (such as radar and communications).
Below are results for the outstanding demos:
In the Deep Cognitive Sparse Arrays for Automotive Radars demo results we can see that by Comparing the DOA estimation performance of three sub-arrays, the proposed CNN approach effectively selects the best sub-array for a large range of SNR and it provides effective performance as compared to random selection. Especially, the CNN-based method has 6dB lower RMSE compared with random selection in the case of MUSIC estimation results.
In the Joint Radar-Communications System demo results, it is observed that the proposed JRC system can obtain higher resolution in radar subsystem and get a better communication performance compared with the fixed allocation scheme.
In the Sub-Nyquist Radar with Distorted Pulse Shape demo results, it is observed that the performance of the proposed single-input multi-output approach is comparable to that of the sub-Nyquist radar. Both the radars operate at 10 times below the Nyquist rate.
Address (URL) of the project's public website https://www.wisdom.weizmann.ac.il/~yonina/YoninaEldar/index.html
The outcome of the research has been implemented in hardware and software demo systems, showing the advantages over traditional approach. These demos have been presented world wide in a large series of conferences.
Areas of interest:
Radar
Communication
Ultrasound
Microscopy
Our results include more efficient and refined algorithms for the above list, based on Deep Learning which we have begun intergrating into our methods. This will provide more effective and faster algorithms we stated above. We are also improving our hardware implementations and integrating them to larger system-wide operations that include several developments combined (such as radar and communications).
Below are results for the outstanding demos:
In the Deep Cognitive Sparse Arrays for Automotive Radars demo results we can see that by Comparing the DOA estimation performance of three sub-arrays, the proposed CNN approach effectively selects the best sub-array for a large range of SNR and it provides effective performance as compared to random selection. Especially, the CNN-based method has 6dB lower RMSE compared with random selection in the case of MUSIC estimation results.
In the Joint Radar-Communications System demo results, it is observed that the proposed JRC system can obtain higher resolution in radar subsystem and get a better communication performance compared with the fixed allocation scheme.
In the Sub-Nyquist Radar with Distorted Pulse Shape demo results, it is observed that the performance of the proposed single-input multi-output approach is comparable to that of the sub-Nyquist radar. Both the radars operate at 10 times below the Nyquist rate.
Address (URL) of the project's public website https://www.wisdom.weizmann.ac.il/~yonina/YoninaEldar/index.html