Recent advancements in deep neural networks have shown remarkable improvements in image quality during the demosaicking process, surpassing conventional algorithms. However, these deep neural network techniques are often characterized by heavy computational requirements, rendering them unsuitable for deployment on resource-constrained platforms. This presents a critical challenge in the field of image demosaicking: while deep learning approaches excel in enhancing image quality, their computational intensity poses a significant hindrance to their wider adoption. Consequently, there is a pressing need for methodologies that can strike an optimal balance between achieving superior image quality and maintaining computational efficiency. In this work, we propose a new deep framework, hyper-prior dependent demosaic neural network, HPDNet that utilizes the significant concepts of the conventional algorithm and the characteristic of image sensor data. We designed the network that exploits three concepts, those are pixel gradient prior attention, phase separation, and multi-level sparse and dense feature extraction. We designed the deep neural network that extracts the optimal gradient prior and the multi-level extracted features are fused and attended by gradient prior. It can fully utilize spatially variant information. Also to make the network deployed in the mobile platform, we devised self-pruned image convolution that adopts image filter characteristic and reduce computations. Experiments show that proposed network outperforms SOTA demosaic networks both in terms of image quality and computation.

We address the problem of developing an accurate but efficient antispoofing (AS) approach in fingerprint biometric recognition on a smart card. To meet low-power constraints for smartcards, we propose a simple convolutional neural network-based architecture and dedicated hardware to handle the problem. To design a simple but well generalized algorithm, we utilize local patches that focus on fingerprint minutiaes and the generalization and accuracy of the network are improved by augmenting local patches centered on the direction map that indicates the direction of the fingerprint. To design flexible hardware, we create a simple 2D convolution unit and leverage it iteratively to cover different channel depths. Thanks to direction-map based data augmentation and simple dedicated hardware, biometric smart cards can protect against spoofing attacks in an accurate yet low-power way with limited number of data.

Multi-pixel image sensors such as Tetra-Cell image sensor have been developed in recent years, and its pixel size is becoming smaller with high resolution. As the amount of pixel data being transmitted has increased, it consumes significantly higher power in the imaging system. Although many image compression algorithms have been studied to reduce the amount of pixel data, it comes with high-complexity, focuses on Bayer pattern, and even has challenges of geometrical distances on Tetra-Cell pattern. In this paper, we propose Multi-Pixel Compression (MPC) which can provide high image quality as well as high compression ratio on Tetra-Cell pattern. We also propose low-power imaging system with MPC. In the proposed imaging system, the power consumption of MIPI transmission and DRAM access can be reduced by 75%. The proposed MPC architecture is implemented as ASIC, its encoder design achieves energy efficiency 1.94 Gb/s/mW and 1.29 Gb/s/mW on Bayer and Tetra-Cell pattern image, respectively, and it achieves 57.11dB PSNR, thus outperforming related previous work.

Deep neural networks have recently advanced state-of-the-art in motion deblurring. However, non-uniform non-blind image deblurring has not been studied in depth. State-of-the-art methods shows improvement over conventional algorithms, but they are still not feasible for mobile deployment. Having informative prior information could improve performance of non-uniform deblurring. In this work, we propose a new deep framework that allows extracting spatially variant latent feature to Distortion Prior map from a pair of calibration sharp-blur images, without having to capture or model training dataset. We propose to use multi-scale Distortion Prior map that can fully utilize spatially variant information in the further restoration via multi-scale attention mechanism. Unlike prior art, we use image pyramid at decoder side, by fusing its fine level with coarse level of feature map via level attention and by injecting Distortion Prior at various resolution levels. Experiments show that proposed network outperforms state-of-the-art deblur networks both in terms of image quality and inference time. We demonstrate that proposed framework can successfully deblur non-uniform, non-blind applications, such as defocus blur removal. Being computationally efficient, it is feasible for mobile deployment.

Tele-wide camera system with different Field of View (FoV) lenses becomes very popular in recent mobile devices. Usually it is difficult to obtain full-FoV depth based on traditional stereo-matching methods. Pure Deep Neural Network (DNN) based depth estimation methods can obtain full-FoV depth, but have low robustness for scenarios which are not covered by training dataset. In this paper, to address the above problems we propose a hierarchical hourglass network for robust full-FoV depth estimation in tele-wide camera system, which combines the robustness of traditional stereo-matching methods with the accuracy of DNN. More specifically, the proposed network comprises three major modules: single image depth prediction module infers initial depth from input color image, depth propagation module propagates traditional stereo-matching tele-FoV depth to surrounding regions, and depth combination module fuses the initial depth with the propagated depth to generate final output. Each of these modules employs an hourglass model, which is a kind of encoder-decoder structure with skip connections. Experimental results compared with state-of-the-art depth estimation methods demonstrate that our method not only produces robust and better subjective depth quality on wild test images, but also obtains better quantitative results on standard datasets.

This paper explores the blind decoding of control channels for obtaining other user identities in 3GPP specification, such as high-speed packet access and long-term evolution. The reliable decoding of control channels with user identities is crucial to mitigate inter-cell interference as well as multi-user interference. This paper exploits a method of user identity filtering followed by a method of user identity detection based on the traffic persistency, which is common to all standards. Hence, the proposed methods are applicable to all the standards regulated by 3GPP specification. In particular, this paper analyzes the proposed other user identity detection algorithm under the random coding. Simulation results show that the proposed method is reliable even at low SNRs and is also aligned with the analysis.

The signal regeneration method (SRE) and associated adaptive algorithms are proposed for fast fading channels. The adaptive algorithms such as the leaky least mean square (LLMS) algorithm assisted by the SRE method provide the signal-to-interference-noise-ratio (SINR) performance comparable to the linear minimum mean square error (LMMSE) equalizer, with low numerical complexity and low memory requirements. The proposed SRE-LLMS algorithm is analyzed based on the independence assumption, and shown to provide low excessive mean square error (EMSE) compared to the LMS algorithm without incurring a bias in its estimate. The simulation results for the wireless channel models such as VA30 and VA120 are presented to validate the analysis.

In this letter, reduced complexity self-tuning algorithms are proposed using simplified parameter updating procedures. Convergence analysis based on the independence assumption and the ordinary differential equation (ODE) method shows that the tuning parameter of the proposed algorithm attains the same limit as the conventional self-tuning adaptive algorithm. Simulations are carried out for channel estimation to support the analysis and performance of the proposed algorithms.

In this letter, pilot design and channel estimation are discussed for orthogonal frequency-division multiplexing (OFDM) systems with guard subcarriers. First, we investigate the effects of guard band on channel estimation errors. From this, we propose pilot placement having a maximum distance between adjacent pilots except for the guard band, and show that it achieves minimum channel estimation errors among partially equispaced pilots using equivalence of the Toeplitz and circulant matrices. Also, an efficient channel estimator is developed by introducing an extended channel and its finite impulse response (FIR) approximation to overcome high numerical complexity caused by the presence of guard subcarriers and the use of a large number of subcarriers. Simulation results are presented for OFDM and orthogonal frequency division multiple access (OFDMA) systems consistent with IEEE 802.16a standards.

In this correspondence, blind channel estimators exploiting finite alphabet constraints are discussed for orthogonal frequency-division multiplexing (OFDM) systems. Considering the channel and data jointly, a joint maximum-likelihood (JML) algorithm is described, along with identifiability conditions in the noise-free case. This approach enables development of general identifiability conditions for the minimum-distance (MD) finite alphabet blind algorithm of Zhou and Giannakis. Both the JML and MD algorithms suffer from high numerical complexity, as they rely on exhaustive search methods to resolve a large number of ambiguities. We present a substantially more efficient blind algorithm, the reduced complexity minimum distance (RMD) algorithm, by exploiting properties of the assumed finite-length impulse response (FIR) channel. The RMD algorithm exploits constraints on the unwrapped phase of FIR systems and results in significant reductions in numerical complexity over existing methods. In many cases, the RMD approach is able to completely eliminate the exhaustive search of the JML and MD approaches, while providing channel estimates of the same quality.

We consider the problem of joint equalization and decoding, using the method of turbo equalization originally developed by Douillard, et al. [3]. In its original form, turbo-equalization requires accurate knowledge of the channel at the receiver. We propose a receiver structure, based on a soft-input Kalman channel estimator, that can operate effectively without accurate channel knowledge and without training data. The resulting joint channel and data estimator is shown to outperform standard turbo equalization based on moderate-length training data.

Variable forgetting factor linear least squares (VFFLLS) algorithm is presented in order to improve the tracking capability of the channel estimation. Compared to the already-existing algorithm-exponentially windowed recursive least squares (EW-RLS) with the optimal forgetting factor, this method makes remarkable improvement in a fast fading environment. The effects of channel parameter-signal to noise ratio, fading rate, are tested in computer simulations with the given channel model. The performance of each algorithm is assessed in terms of the mean-square-identification-error (MSIE).