Side channel analysis based on non-profiling singular spectrum analysis

Side-Channel Analysis (SCA) plays a crucial role in hardware security evaluation. However, side-channel acquisitions (a.k.a. traces) usually contain noises that often impose negative effects on key-recovery efficiency. In this paper, we propose convolutional denoising autoencoder (CDAE) for noise reduction in SCA. CDAE is composed of multiple layers of convolution operators, learning an end-to-end mapping from noisy traces to clean traces by minimizing the \(\ell _2\) loss of noisy-clean trace pairs. The convolutional layers capture the abstraction of the traces while eliminating noises. We argue that CDAE is very suitable for profiled SCA especially when the attacker has a large amount of traces in the offline profiling phase. Once the network training is done, our denoising network can be applied to individual new noisy traces for the attacker to launch online attacks. To validate the effectiveness of our method, we train CDAE to denoise traces and then perform Template Attacks (TA) in three high noise jamming scenarios, including unprotected (GPU and FPGA based) and protected (MCU based) AES implementations. Our method can significantly outperform the state-of-the-art Singular Spectrum Analysis (SSA) denoising method on both information theoretic metrics and security metrics. Results show that CDAE achieves at least \(\sim 4\times \) Signal-to-Noise Ratio (SNR) gain, thus TA with denoising preprocessing requires at most 50% of the traces in the attack phase.

The complexity of the cancer problem domain presents challenges not only to the medical analysis systems tasked with its analysis, but also to the users of such systems. While it is desirable to assist users in operating these medical analysis systems, prior groundwork is required before this can be achieved, such as recognising patterns in the way users create certain analyses within these systems. In this paper, we use machine learning algorithms to analyse user behaviour patterns and attempt to predict the next user interaction within the CARESS medical analysis system. Since an appropriate pre-processing scheme is essential for the performance of these algorithms, we propose the usage of a Natural Language Processing (NLP)- inspired approach to preserve some semantic cohesion of the mostly categorical features of these user interactions. Furthermore, we propose to use a sliding window that contains information about the latest user interactions in combination with Latent Dirichlet Allocation (LDA) to extract a latent topic from these last interactions and use it as additional input to the machine learning models. We compare this pre-processing scheme with other approaches that utilise one-hot encoding and feature hashing. The results of our experiments show that the sliding window LDA scheme is a promising solution, that performs better for our use case than the other evaluated pre-processing schemes. Overall, our results provide an important piece for further research and development in the area of assisting users in operating analysis systems in complex problem domains.

In ocean acoustic fields, extracting the normal-mode interference spectrum (NMIS) from the received sound intensity spectrum (SIS) plays an important role in waveguide-invariant estimation and underwater source ranging. However, the received SIS often has a low signal-to-noise ratio (SNR) owing to ocean ambient noise and the limitations of the received equipment. This can lead to significant performance degradation for the traditional methods of extracting NMIS at low SNR conditions. To address this issue, a new deep neural network model called SSANet is proposed to obtain NMIS based on unrolling the traditional singular spectrum analysis (SSA) algorithm. First, the steps of embedding and singular value decomposition (SVD) in SSA is achieved by the convolutional network. Second, the grouping step of the SSA is simulated using the matrix multiply weight layer, ReLU layer, point multiply weight layer and matrix multiply weight layer. Third, the diagonal averaging step was implemented using a fully connected network. Simulation results in canonical ocean waveguide environments demonstrate that SSANet outperforms other traditional methods such as Fourier transform (FT), multiple signal classification (MUSIC), and SSA in terms of root mean square error, mean absolute error, and extraction performance.

Electroencephalogram (EEG) signals are crucial to Brain-Computer Interfacing (BCI). However, these are vulnerable to a variety of unintended artifacts that could negatively impact the precise brain function assessment. This paper provides a new algorithm to eliminate Electrical Shift and Linear Trend artifact (ESLT) in EEG using Singular Spectrum Analysis (SSA) and Enhanced local Polynomial (LP) Approximation-based Total Variation (EPATV). The contaminated single channel EEG is subdivided into multiple bands of frequency components by SSA. In order to acquire all LP and TV components, EPATV filtering is applied over the contaminated component frequency band. Filtered sub-signal is collected by subtracting both the LP and TV components from the component contaminated frequency band. Then, the addition of filtered sub-signal and remaining SSA frequency band components yield the final denoised EEG signal. The effectiveness of the proposed method in this paper is evaluated using the data obtained from three databases and compared with the existing methods. From the extensive simulation results, it is inferred that the algorithm discussed in the paper is effective when compared the existing methods, exhibiting a highest averaged Correlation Coefficient (CC) of 0.9534, averaged Signal to Noise Ratio (SNR) of 10.2208dB, lowest averaged Relative Root Mean Square Error (RRMSE) value 0.2787 and averaged Mean absolute Error (MAE) in α band value of 0.0557. The algorithm presented in this paper may be a viable choice for extracting ESLT artifact from a small streaming section of the EEG without requirement of the initial calibration or enormous EEG data.

Simple SummaryNeurosurgical decisions for patients with glioblastoma depend on visual inspection of a preoperative MR scan to determine the tumor characteristics. To avoid subjective estimates and manual tumor delineation, automatic methods and standard reporting are necessary. We compared and extensively assessed the performances of two deep learning architectures on the task of automatic tumor segmentation. A total of 1887 patients from 14 institutions, manually delineated by a human rater, were compared to automated segmentations generated by neural networks. The automated segmentations were in excellent agreement with the manual segmentations, and external validity, as well as generalizability were demonstrated. Together with automatic tumor feature computation and standardized reporting, our Glioblastoma Surgery Imaging Reporting And Data System (GSI-RADS) exhibited the potential for more accurate data-driven clinical decisions. The trained models and software are open-source and open-access, enabling comparisons among surgical cohorts, multicenter trials, and patient registries.For patients with presumed glioblastoma, essential tumor characteristics are determined from preoperative MR images to optimize the treatment strategy. This procedure is time-consuming and subjective, if performed by crude eyeballing or manually. The standardized GSI-RADS aims to provide neurosurgeons with automatic tumor segmentations to extract tumor features rapidly and objectively. In this study, we improved automatic tumor segmentation and compared the agreement with manual raters, describe the technical details of the different components of GSI-RADS, and determined their speed. Two recent neural network architectures were considered for the segmentation task: nnU-Net and AGU-Net. Two preprocessing schemes were introduced to investigate the tradeoff between performance and processing speed. A summarized description of the tumor feature extraction and standardized reporting process is included. The trained architectures for automatic segmentation and the code for computing the standardized report are distributed as open-source and as open-access software. Validation studies were performed on a dataset of 1594 gadolinium-enhanced T1-weighted MRI volumes from 13 hospitals and 293 T1-weighted MRI volumes from the BraTS challenge. The glioblastoma tumor core segmentation reached a Dice score slightly below 90%, a patientwise F1-score close to 99%, and a 95th percentile Hausdorff distance slightly below mm on average with either architecture and the heavy preprocessing scheme. A patient MRI volume can be segmented in less than one minute, and a standardized report can be generated in up to five minutes. The proposed GSI-RADS software showed robust performance on a large collection of MRI volumes from various hospitals and generated results within a reasonable runtime.

Security at low cost is an important factor for cryptographic hardware implementations. Unfortunately, the security of cryptographic implementations is threatened by Side Channel Analysis (SCA). SCA attempts to discover the secret key of a device by exploiting implementation characteristics and bypassing the algorithm's mathematical security. Differential Power Analysis (DPA) is a type of SCA, which exploits the device's power consumption characteristics. Several countermeasures to DPA have been proposed, however, all of them increase security at the cost of increased area which in-turn leads to increased power consumption and reduced throughput. FPGAs are popular due to their reconfigurability, lower development cost, off-the-shelf availability and shorter time to market. Block RAMs (BRAM) are large memories in FPGAs that are commonly used as ROM, FIFO, Look-up tables, etc. In this paper we explore the DPA resistance of BRAMs in Xilinx FPGAs and verify if their usage can improve the security. The results of our Advanced Encryption Standard (AES) implementations show that using BRAMs alone can improve the security over a look-up table (LUT) only design 9 times. Applying Separated Dynamic Differential Logic (SDDL) for FPGAs, a countermeasure against DPA, to this design doubles the security again leading to an 18 fold increase over the unprotected LUT design.

USB keyboards are commonly used as computer input devices and inevitably generate electromagnetic (EM) leakage signals during their operation, which carry input information. However, due to the weak energy of a keyboard’s EM signal and the small amount of effective information, the received leakage signal is often characterized by a low signal-to-noise ratio (SNR). This low SNR affects the subsequent detection and restoration of the information. In order to solve this problem, this paper proposes a denoising method for USB keyboard EM leakage signals and designs a self-attentive denoising adversarial network (SADAN) based on generative adversarial networks (GANs). The denoiser continuously enhances the denoising ability during the generative adversarial process, and the self-attention mechanism enables it to better learn the dependencies of the keyboard EM leak signal sequences, modelling the long-range relationships between the sequence sample points and reducing the impact of the number of network layers on the relationship acquisition. The method achieves noise suppression in the keyboard leakage signal, improving its SNR while preserving the effective information in the leakage signal and finally obtaining a denoised leakage signal that can be effectively restored to the information.

Sensitive cryptographic information, e.g. AES secret keys, can be extracted from the electromagnetic (EM) leakages unintentionally emitted by a device using techniques such as Correlation Electromagnetic Analysis (CEMA). In this paper, we introduce Correlation Optimization (CO), a novel approach that improves CEMA attacks by formulating the selection of useful EM leakage samples in a trace as a machine learning optimization problem. To this end, we propose the correlation loss function, which aims to maximize the Pearson correlation between a set of EM traces and the true AES key during training. We show that CO works with high-dimensional and noisy traces, regardless of time-domain trace alignment and without requiring prior knowledge of the power consumption characteristics of the cryptographic hardware. We evaluate our approach using the ASCAD benchmark dataset and a custom dataset of EM leakages from an Arduino Duemilanove, captured with a USRP B200 SDR. Our results indicate that the masked AES implementation used in all three ASCAD datasets can be broken with a shallow Multilayer Perceptron model, whilst requiring only 1,000 test traces on average. A similar methodology was employed to break the unprotected AES implementation from our custom dataset, using 22,000 unaligned and unfiltered test traces.

Short-term wind speed forecasting based on long short-term memory and improved BP neural network

Distributed acoustic sensing (DAS) is a new tool with low cost, sensitive signal capture, and complete coverage for vertical seismic profile (VSP) acquisition. Although DAS has obvious advantages over geophones, some weaknesses may limit its application. The main challenge is that DAS is polluted by various types of noise, including optical abnormal noise, random background noise, fading noise, and so on. To suppress these novel noises, we developed a new denoising neural network based on singular spectrum analysis—multichannel singular value decomposition denoising convolutional neural network (SVDDCNN). The network can simultaneously extract data features from singular spectrum instead of the time domain, which can represent geophysical features more accurately and help separate signals from noises. Second, a multichannel input layer is designed, and the input is decomposed into three subspaces by singular spectrum analysis, which provides records of different signal-to-noise ratios (SNRs) for training and improves generalization ability of the network. Third, to enhance the quality of the data set, we added the noise subspace records removed by SVD into the training set to provide various forms of noise with different singular spectra. Both synthetic and field examples show that our network has achieved impressive denoising of DAS VSP and demonstrated competitive performance compared with other methods. Furthermore, the structure similarity (SSIM) map is introduced to evaluate the signal leakage by calculating the similarity between the denoised record and the removed noise record. The lowest SSIM index of the proposed network indicated superior signal preservation ability.

Life signal mainly refers to the respiration signal and heartbeat signal, which is an objective index of the organism’s health status. Therefore, the detection of life signal is of great clinical and practical significance. In recent years, life signal detection with terahertz radar is a keen research topic owing to terahertz wave’s superior Doppler performances. In this paper, the life-signal detection method based on Singular Spectrum Analysis (SSA) is proposed. According to the simulation carried out in the Gaussian clutter background with a signal-to-noise ratio (SNR) ranging from 10 dB to -20dB, it shows that the proposed method works better than Empirical Mode Decomposition (EMD) under low SNR. A 0.33 THz radar system and the experiment on life-signal detection with it are presented. The results demonstrate the method’s performance, which could be applied to the remote monitoring of life signal of a patient in the following application, and bring new ideas to related research.

This paper investigates the simultaneous wireless information and power transfer (SWIPT) for network-coded two-way relay network from an information-theoretic perspective, where two sources exchange information via an SWIPT-aware energy harvesting (EH)relay. We present a power splitting (PS)-based two-way relaying (PS-TWR) protocol by employing the PS receiver architecture. To explore the system sum rate limit with data rate fairness, an optimization problem under total power constraint is formulated. Then, some explicit solutions are derived for the problem. Numerical results show that due to the path loss effect on energy transfer, with the same total available power, PS-TWR losses some system performance compared with traditional non-EH two-way relaying, where at relatively low and relatively high signal-to-noise ratio (SNR), the performance loss is relatively small. Another observations is that, in relatively high SNR regime, PS-TWR out-performs time switching-based two-way re-laying (TS-TWR) while in relatively low SNR regime TS-TWR outperforms PS-TWR. It is also shown that with individual available power at the two sources, PS-TWR outperforms TS-TWR in both relatively low and high SNR regimes.

MISO (Multiple-Input Single-Output) similar architecture of most terrestrial wireless network networks instead of more and more scientists striving to apply MISO technology to satellite technology, it can be used to achieve a lower rate of application bit error and risk of complications, while managing increased power boost technology. In mitigating the phenomenon increasingly crowded networks, sites, and increasingly strained sources and frequency of orbital interacting. That channel electricity, bit error rate, and single-band dual-polarized probabilities of MISO outage communication systems are evaluated in the paper. In the first step, when the XPD (Cross-Polarization Discrimination) antenna in a certain organism is greater than 1. With the XPD change, the channel throughput increases in scale linearly. Second, under BPSK modulation, it analyses the bit error rate BER of the specification. The design incorporates a sufficiently low (BER) when the SNR is substantial; analysis of the frequency of outage of a distributed MISO system demonstrated when the signal to noise ratio (SNR) is massive, the handset would maintain the prospect of an outage low enough. Assessment of the spectral range of satellite communication, in the analysis process, the criteria of the rice channel are applied to the channel, review excluding certain traits of the dual-polarized satellite MISO. The public would provide great feedback for the productivity of the future of MISO satellite technology. In a previous paper, the author designed a channel modeling dual-band for satellite communication using the MIMO technique. Using this technique, the system cannot achieve greater performance and at the same time using a dual-band will decrease the system's capability. To decrease the above issues, the paper introduces a new method called the MISO system. Using the MISO technique, the system can gain high performance and the system will get greater bandwidth, BER, and SNR. The advantage of using single-band dual-polarized is, uplink and a downlink frequency of the satellite can be calculated very easily and accurately.

With the prevalence of the Internet of Things (IoT) and microcontrollers (MCUs), the security issues of IoT and MCUs have become increasingly important. Side-channel analysis (SCA) is a major threat to such problems. One of the non-invasive SCAs is through electromagnetic information leakage (EM-leak) analysis. The author has developed a machine-instruction-level EM-leak analysis by neural network (NN) model. The NN model analysis need a large dataset for training and validation. And the dataset should be complete and sufficient. The dataset sufficiency can be achieved by acquiring more data from the already proposed EM-leak measurement platform. However, the completeness issue becomes a challenge due to the pipelined architecture in the target MCU. In this paper, the completeness issue of a NN model dataset for the EM-leak analysis is addressed. The issue is to be solved in several steps. First, it is simplified to a two-stage pipelined MCU individual EM-leak analysis sequence problem (2-EMAseq). The problem is then proved that at least one optimal solution exists. And an algorithm is proposed based on the proof to find an optimal solution (a complete EM-leak NN dataset). Experiments show that the proposed algorithm can find an optimal solution. The contributions of this paper include: it successfully reduces a real problem to a 2-EMAseq description, proves that the 2-EMAseq problem has at least one optimal solution, uses this proof to develop an algorithm, and uses this algorithm to find a complete dataset is the target two-stage pipelined MCU dataset for NN model training/validation. Currently, the algorithm can only generate the optimal EM-dataset for two-stage pipelined MCUs. However, there are many MCUs with 4~8 pipeline stages. Whether there is an optimal solution when the stages are more than two is still an open question. And it needs to find proofs or derive heuristics in the future for such MCUs.

Side-channel analysis (SCA) is launched by exploiting the information leaking from the implementation a cryptographic algorithm, e.g., power consumption information. Recently, deep learning-based SCA techniques have also facilitated SCA against software and hardware implementations of various cryptographic algorithms. In this work, we perform SCA using various deep learning (DL) techniques such as Multi-layered Perceptron (MLP), Convolutional Neural Network (CNN), and Recurrent Neural Network (RNN) on the datasets collected from hardware and software platforms. The objective of this work is to identify the performance of DL techniques in performing SCA for secret key recovery and finding out the best settings for the model to optimize the attack performance in terms of on computation time and SCA efficiency. In our study, we have focused on two open-source AES-128 encryption algorithm databases, ASCAD and DPA contest v2 (DPAv2), where ASCAD database consists of the power traces captured from a software implementation of the AES and DPAv2 database consists of the power traces captured a hardware implementation of the AES. For the first time, we applied hyperparameter tuning with Bayesian Optimization and distributed computing on ASCAD database and we investigated the impact of MLP and RNN along with the distributed computing and hyperparameter tuning with Bayesian optimization on DPAv2 database. Our results show that the CNNs are the best models for performing the attack on software implementation while MLPs are the best for attacking hardware implementation of cryptographic algorithms.