Mohammad Fakharzadeh

In this article, sparse sampling in the frequency domain for increasing the speed of the wideband microwave imaging systems is investigated. First, a thorough analysis on imaging theory and range ambiguity is conducted. A new metric called array-frequency response (AFR), is introduced for assessing the performance of the 2-D imaging systems based on both the array configuration and frequency sampling scheme. Two algorithms, namely iterative frequency selection (IFS) and gradient-descent frequency optimization (GDFO), are introduced for selection of effective frequencies. Both algorithms are intended to reduce the number of frequency samples, while maintaining the image quality. A wideband (10–20 GHz) 2-D imaging system with $40 \times 40$ antenna elements is designed and successfully implemented to verify the proposed algorithms. It is shown that the IFS algorithm is able to reduce the number of frequencies by a factor of 3.5 while the reconstructed image quality is 95% of the reference image reconstructed with all frequency samples, based on SSIM metric.

In this paper, the problem of beamforming and maximizing the phased array gain for practical phase shifters is addressed using a deep learning approach. For an ideal phase shifter, the amplitude is constant and the output phase changes linearly with the control voltage. However, the practical phase shifters show a nonlinear behavior in both amplitude and phase with respect to the control voltage. The main objective of this paper is to design a deep neural network, which estimates the optimum voltage values of the phase shifter using the noisy data of the receiver antenna array. The proposed approach is general and extendible to any array geometry. Usually, the quality of the trained model is affected by large variations in the voltage values for certain adjacent Directions of Arrival (DOAs). To overcome this issue, we introduce the robust mapping technique, which avoids sudden changes in voltage values during the training process. The amplitude and phase data of a typical reflective-type phase shifter is measured for different control voltage values, to be used in the training process. The suggested algorithm is applied to a receiver array of four antennas with reflective type phase shifters. The maximum array gain for different DOAs are very close to the optimum results showing the superior performance of the proposed method. Compared to the existing methods, the RMSE of the array gain decreased to 0.0283. It is shown that the proposed method is more resilient to noise and performs substantially better than other methods in low SNR scenarios.

In millimeter-wave (MMW) imaging, the objects of interest are oftentimes modeled as 2D binary (black and white) shapes with white pixels representing the reflecting interior of the object. However, due to propagation of the scattered waves, the continuous-domain binary images are convolved with a so-called point-spread function (PSF) before being digitized by means of sampling. As the 2D PSF is both non-separable and non-vanishing in the case of MMW imaging, exact recovery is quite complicated. In this paper, we propose a deep learning approach for image reconstruction. We should highlight that the wave scatterings are suitably represented with complex-valued quantities, while standard deep neural networks (DNN) accept real-valued inputs. To overcome this challenge, we separate the real and imaginary parts as if we had two imaging modalities and concatenate them to form a real-valued input with larger size. Fortunately, the network automatically learns how to combine the mutual information between these modalities to reconstruct the final image. Among the advantages of the proposed method are improved robustness against additive noise and mismatch errors of imaging frequency and object to antenna distance; indeed, the method works well in wide-band imaging scenarios over a wide range of object to antenna distances even in presence of high noise levels without requiring a separate calibration stage. We test the method with synthetic data simulated with software as well as real recordings in the laboratory.

This paper reports a new device for electrocardiogram (ECG) signal monitoring and software for signal analysis and artificial intelligence (AI) assisted diagnosis. The hardware mitigates the signal loss common in previous products by enhancing the ergonomy, flexibility, and battery life. The power efficiency is optimized by design using switching converters, ultra-low-power components, and efficient signal processing. It enables 14-day of uninterrupted ECG monitoring and connectivity with a smartphone and microSD card storage. The software is implemented in Android app and web-based platforms via Internet of Things (IoT). This component provides cloud-based and local storage and uses AI for arrhythmia detection. The arrhythmia detection algorithm shows 98.7% accuracy using Artificial Neural Network and K-Nearest Neighbors methods, and 98.1% using Decision Tree method on test data set.

This paper reports the development of a new smart ECG monitoring system, consisting of the related hardware, firmware, and IoT-based web service for AI assisted arrhythmia detection and a complementary Android application for data streaming. The hardware aspect of this research proposes an ultra-low power patch sampling ECG data at 256 samples/s with 16-bit resolution. The battery life of the device is two weeks per charging, which alongside the flexible and slim (193.7 mm × 62.4 mm × 8.6 mm) and lightweight (43 g) allows the user to continue real life activities while the real-time monitoring is being done without interruption. The power management is achieved through the usage of switching converters, ultra-low power component choice as well as intermittent usage of them through firmware optimization. A novel data encoding method is also proposed to allow compression of data and lower the runtime. The software aspect, in addition to the web ECG analysis platform and the Android streaming and monitoring application, provides an arrhythmia detection service. The key innovations in this regard are the usage of a set of new factors in determining arrhythmia that grants higher accuracy while retaining the detection near-real-time. The arrhythmia detection algorithm shows 98.7% accuracy using Artificial Neural Network and K-Nearest Neighbors methods, and 98.1% using Decision Tree method on test data set.

This paper adopts a holistic approach to stress detection issues in software and hardware phases and aims to develop and evaluate a specific low-power and low-cost sensor using physiological signals. First, a stress detection model is presented using a public data set, where four types of signals, temperature, respiration, electrocardiogram (ECG), and electrodermal activity (EDA), are processed to extract 65 features. Using Kruskal-Wallis analysis, it is shown that 43 out of 65 features demonstrate a significant difference between stress and relaxed states. K nearest neighbor (KNN) algorithm is implemented to distinguish these states, which yields a classification accuracy of 96.0±2.4%. It is concluded that ECG and EDA signals are sufficient to detect stress with high accuracy while the system has fewer sensors and consumes lower power. Second, a novel sensor is developed to collect ECG and EDA signals from eighteen healthy participants aged 16 to 40 years. They were exposed to stress during an arithmetic mental task and Stroop Color-Word Test. This sensor has a battery life of 70 hours and can detect stress with 94.4±2.5% accuracy. The proposed software and hardware can potentially be used in a wearable device for continuous stress detection.

This communication addresses the focusing problem in the millimeter-wave imaging systems. We categorize the focusing problem into the frequency focusing for wideband systems and the range focusing for narrowband systems. In an out-of-focus wideband system, a shifted shadow of the object is present in the reconstruction, whereas for a range out of the focused system, the recovered images are blurred. To overcome these issues, first, we theoretically show that the defocusing variations for both categories are bounded. Then, we present a universal formulation for focusing problem, which covers both wideband and narrowband systems. As the true focused images are sharp at the boundaries of the objects, our strategy for solving the problem is to maximize a defined sharpness metric. Moreover, we propose an autofocusing zero-knowledge algorithm, which concerns with maximizing the sharpness metric from an unknown object, while the exact gradient of the cost function is unknown. The proposed method is suitable for practical applications since it is simple, fast, and computationally efficient. The simulation results on synthetic and measured data are promising and support our claims that the proposed method increases the quality of the reconstruction.

In this paper, we propose a software-hardware system to improve the quality of image recovery in millimeter wave (MMW) imaging systems, which aims to benefit from co- and cross-polarization scattering from the target. Dual polarized antenna array structures are proposed consisting of two orthogonal subarrays with the same phase center, where each subarray is comprised of 2 × 1 or 4 × 1 rectangular edge fed patches as well as the optimized ground defects. These compact structures, which support radiation of two orthogonal linear polarizations, are designed on a single-layer board using crossover technique. The dual-polarized scattering from the target is captured by the orthogonal subarrays. Furthermore, we introduce a method to combine data obtained from measured dual (co- and cross-)polarized fields to improve the recovery of edge information, and increase image quality. The proposed single layer designs have been fabricated and measured, which exhibit high port isolation and high polarization purity. The cross polarization level is at least 22 dB and 25 dB lower than the co-polarization pattern for 2 × 1 and 4 × 1 arrays, respectively. The measured 10-dB impedance bandwidth of the 2 × 1 and 4 × 1 subarrays range from 27 to 29.3 GHz and from 27 to 29.5 GHz, respectively. Moreover, the radiation gain of the 2 × 1 and 4 × 1 subarrays are 9.7 dBi and 12 dBi, respectively. The simulated and measured imaging results show that the cross polarized data preserve the edge information. Moreover, by using the proposed data combination technique, the quality of the recovered image is improved.

This article presents a novel dual-band transmission line (TL), consisting of a multibranch stub, which can provide equal unwrapped phases at two different frequencies. The frequency ratio can be close to one without any significant degradation in the device performance. Besides, the bandwidth of the proposed line can be manipulated at both frequencies by adjusting the design parameters. The behavior of the structure over the frequency bands is investigated theoretically, and the design procedure and relations are derived. The simulation and measurement results of a quarter-wavelength open-ended stub and a branch-line coupler (BLC) are presented and compared to verify the performance of the proposed dual-band TL. It is shown that the first structure has a simulated and measured insertion phase of 90° at 2 and 3 GHz. The BLC is designed for 4G and 5G applications with two center frequencies of 3.52 and 4.78 GHz, resulting in a frequency ratio of 1.36. The measured insertion loss values at these frequencies are 3.8 and 3.2 dB, the coupled port loss values are 3.5 and 3.1 dB, and the output phase differences are 91.7° and 92.8°, respectively.