Projects
Proposals
Radar
Project Title:

Cognitive Foliage Penetration Radar (FPR)

Abstract:

The objective of this project is to improve detection of low Doppler targets in the ground-based Foliage Penetration Radar. The native radar system transmits a train of pulses and suffers from interference in the UHF band. We propose cognitive transmission to mitigate the radio-frequency interference and sub-Nyquist sampling to enhance the Doppler resolution to detect slowly moving targets. We focus on the transmission of sparse stepped-frequency continuous waveform (s-SFW) to enable the cognition and then develop sub-Nyquist processing for such a waveform.

Project description:

A foliage penetration (FOPEN) radar detects, tracks and images targets such as human intruders in a dense forest environment. The radar transmission should be able to penetrate the foliage without significant signal attenuation and the received signal should have sufficient backscatter from targets (vehicles and humans) to achieve good detection performance. In order to meet these criteria, often VHF and UHF frequencies are considered suitable candidates.


Description:

PDF

Supervisor(s):
Requirements:
Signal and systems (essential), Mavlas (essential), Random signals (desirable). MATLAB.
Machine Learning
Project Title:

DEEP LEARNING OF SIGNAL PROCESSING CONCEPTS

Abstract:

Signal processing concerns the analysis, synthesis, and modification of signals.
With the advent of digital computers, in the last six decades, digital signal processing (DSP) algorithms have been developed for several applications such as denoising, detection, estimation, etc. There are certain fundamental building blocks which are, in general, common in such algorithms, such as sampling, fast-Fourier transform, linear filtering, etc.
Recently, a tremendous amount of interest has been shown in solving the problems of denoising, detection, estimation, classification, etc. by applying machine learning algorithms.

Description:

PDF

Supervisor(s):
Requirements:
Signal and systems, Mavlas, Mavla. Environment:Matlab
Communication
Project Title:

Hybrid Analog-Digital Beamforming for Wideband Signal Model

Abstract:

Massive MIMO systems are considered as one the leading enabler of 5G wireless communication. In this technology, the transmitter and receiver are equipped with very large number of antennas. This can potentially allow for higher data rates and better spectral efficiency.

One of the main challenges is massive MIMO system is the hardware complexity.

When considering hundreds of antennas, dedicated RF chain per antenna like in traditional MIMO systems is no longer possible. Hence, it is desirable to reduce the number of RF chains is the system while still benefiting from the large number of antennas.

To this aim, a hybrid analog-digital architecture is suggested, where some of the processing, traditionally performed in the digital domain, are shifted to the analog domain. This technique is call hybrid analog-digital beamforming

Therefore, an efficient design method for the hybrid beamformer is required.

Most of the previous works on this field considered a narrowband signal model. However, a wideband signal model is a more accurate assumption in most cases, especially if the system operates in mmWave technology.

The goal of the project is to generalize a previously suggested framework [1] for hybrid beamformers design to the wideband signal case: to incorporate the wideband signal model in the existing framework and make all the necessary adjustments to the theorems and algorithms.

The project will include research next to matlab implementation

Description:

PDF

Supervisor(s):
Requirements:
matlab, mavlas, computational methods for optimization
Project Title:

Deep Hybrid Analog-Digital Beamforming

Abstract:

Massive MIMO systems are considered as one the leading enabler of 5G wireless communication. In this technology, the transmitter and receiver are equipped with very large number of antennas. This can potentially allow for higher data rates and better spectral efficiency.

One of the main challenges is massive MIMO system is the hardware complexity.

When considering hundreds of antennas, dedicated RF chain per antenna like in traditional MIMO systems is no longer possible. Hence, it is desirable to reduce the number of RF chains is the system while still benefiting from the large number of antennas.

To this aim, a hybrid analog-digital architecture is suggested, where some of the processing, traditionally performed in the digital domain, are shifted to the analog domain. This technique is call hybrid analog-digital beamforming

Therefore, an efficient design method for the hybrid beamformer is required.

The goal of the project is to develop a deep learning framework for hybrid beamformers design: to define a cost function, produce a learning set of efficient beamformers, and develop the deep algorithm to produce such beamformers given a new system setting.

The project will include research next to matlab implementation

Description:

PDF

Supervisor(s):
Requirements:
matlab, mavlas, introduction to machina learning
Project Title:

Unimodular Coding for Hybrid Analog-Digital Beamforming

Abstract:

Massive MIMO systems are considered as one the leading enabler of 5G wireless communication. In this technology, the transmitter and receiver are equipped with very large number of antennas. This can potentially allow for higher data rates and better spectral efficiency.

One of the main challenges is massive MIMO system is the hardware complexity.

When considering hundreds of antennas, dedicated RF chain per antenna like in traditional MIMO systems is no longer possible. Hence, it is desirable to reduce the number of RF chains is the system while still benefiting from the large number of antennas.

A framework for hybrid combiner design was previously developed in the lab [1], in which on of the suggested algorithm is a greedy algorithm that solves a vector problem ((37) in [1]) at each iteration. Currently this problem is solved using a dictionary, but this solution is sub-optimal. It is desirable to find a better solution for this problem as it will boost the algorithm’s performance. One possible approach for solution is using unimodular coding [2].

The goal of the project is to develop a unimodular coding based solution to problem (37) in [1].

The project will include research next to matlab implementation

Description:

PDF

Supervisor(s):
Requirements:
matlab, mavlas, computational methods for optimization
Deep Learning
Project Title:

Deep Learning for Physical Layer Communications

Abstract:

Machine learning (ML) has been widely applied to the upper layers of communication systems for various purposes, such as deployment of cognitive radio and communication network. However, its application to the physical layer, and in particular to channel decoding, is limited due to the curse of dimensionality, which restricts the learning ability in practical setups with large codelengths.

Deep learning (DL) has been recently applied for many fields, such as computer vision and natural language processing, given its expressive capacity and convenient optimization capability. In this project we explore the application of DL physical layer receiver design, focusing on channel decoding. Our goal is to exploit the structure of channel decoders to design learning algorithms suitable for this specific problem, which can be optimized for a specific hardware configuration and channel setup.


Description:

PDF

Supervisor(s):
Requirements:
Matlab, Introduction to Digital Signal Processing, Introduction to Digital Communications.
Project Title:

Sparse 3D Imaging using Deep Learning

Abstract:

Diagnostic sonography allows visualization of body tissues, by radiating them with acoustic energy pulses. As the pulse propagates, echoes are scattered by density and propagation-velocity perturbations in the tissue, and detected by the transducer elements. Averaging the detected signals, after their alignment with appropriate time-varying delays, allows localization of the scattering structures, while improving the signal-to-noise ratio (SNR). The latter process is referred to as beamforming.


Description:

PDF

Supervisor(s):
Requirements:
Signal and systems, Mavlas Environment: MATLAB
Project Title:

Super Resolution of Ultrasound Doppler Imaging using Deep Learning

Abstract:

Doppler ultrasound is a non-invasive and safe modality that is used for the estimation of blood velocities by transmitting high-frequency sound waves (ultrasound) and analyzing the signals reflected from circulating red blood cells. Doppler scans help diagnose many conditions, including: heart valve defects and congenital heart disease, artery occlusions and aneurysms.

Description:

PDF

Supervisor(s):
Requirements:
Signal and systems, Mavlas Environment: MATLAB
Biomedical
Project Title:

sub-sampling of functional MRI data and prediction of cognitive ability based on full ve sub-sampling data

Abstract:

Blood oxygenation level dependent Functional Magnetic Resonance Imaging (BOLD-MRI) is a data acquisition technique that tracks the changes in blood flow in task-related regions. Due to the low data acquisition fMRI sample rate, a relatively larger number of volumes should be acquired in order to receive a meaningful signal. One way to achieve this larger number of volumes is by increasing the scanning time, which, in children, can be very problematic due to their inability to stay still and avoid motion throughout the scan.

Description:

PDF

Supervisor(s):
Requirements:
matlab, mavlas
Project Title:

Machine learning for blood glucose prediction-part2

Abstract:

Project description:

Project description:

Diabetes was recently declared as a worldwide epidemic by the National Health Organization, with hundreds of millions of patients worldwide, and many more in a pre-diabetic state.

Diabetes patients need to monitor their blood glucose levels constantly, and are prone to high (hypercalcemia) and low (hypoglycemia) spikes for blood glucose fluctuations.

One of the promising ways to measure glucose levels is with the Libre system, which attaches to the back of the patients hand.

In this project we will develop a recurrent network to predict blood glucose levels, based on the collected data from the Freestyle Libre.


Description:

PDF

Supervisor(s):
Requirements:
matlab, mavlas
Project Title:

High Resolution Beamforming using Deep Learning

Abstract:

Diagnostic sonography allows visualization of body tissues, by radiating them with acoustic energy pulses. As the pulse propagates, echoes are scattered by density and propagation-velocity perturbations in the tissue, and detected by the transducer elements. Averaging the detected signals, after their alignment with appropriate time-varying delays, allows localization of the scattering structures, while improving the signal-to-noise ratio (SNR). The latter process is referred to as beamforming.


Standard beamforming is a generic process, performed without exploiting the knowledge of the medium being scanned. Often, it results with poor images due to low resolution or technician's movement. This makes it difficult for the doctor to give a reliable diagnosis.

In this project, we aim to develop an organ-based beamformer using deep learning nets. Selecting an area and applying the corresponding learned beamformer will allow the doctor to view the region of interest with high resolution. The project will include the study of the basics ultrasound imaging and designing CNNs.


Description:

PDF

Supervisor(s):
Requirements:
Introduction to Digital Signal Processing (044198) Learning System (046195) Tensor Flow – Advantage
INTERIA Web Design & Development