The fusion of infrared (IR) and visible (VIS) images aims to capture complementary information from diverse sensors, resulting in a fused image that enhances the overall human perception of the scene. However, existing fusion methods face challenges preserving diverse feature information, leading to cross-modal interference, feature degradation, and detail loss in the fused image. To solve the above problems, this paper proposes an image fusion method based on the infrared target mask and bimodal feature extraction strategy, termed IBFusion. Firstly, we define an infrared target mask, employing it to retain crucial information from the source images in the fused result. Additionally, we devise a mixed loss function, encompassing content loss, gradient loss, and structure loss, to ensure the coherence of the fused image with the IR and VIS images. Then, the mask is introduced into the mixed loss function to guide feature extraction and unsupervised network optimization. Secondly, we create a bimodal feature extraction strategy and construct a Dual-channel Multi-scale Feature Extraction Module (DMFEM) to extract thermal target information from the IR image and background texture information from the VIS image. This module retains the complementary information of the two source images. Finally, we use the Feature Fusion Module (FFM) to fuse the features effectively, generating the fusion result. Experiments on three public datasets demonstrate that the fusion results of our method have prominent infrared targets and clear texture details. Both subjective and objective assessments are better than the other twelve advanced algorithms, proving our method's effectiveness.

The Resistive-RAM (RRAM) based deep neural network (DNN) accelerators have shown great potential as they are good at solving matrix-vector multiplication (MVM). However, this computing paradigm does not benefit other NN operations like activation, which may be built upon various transcendental functions and require customized circuit as in current RRAM-based NN accelerators. In this paper, we propose the RRAM-CORDIC algorithm and crossbar design which enable various transcendental activation calculations on a RRAM crossbar just like MVM. By applying encoding and multi-iteration transformation, the RRAM-CORDIC can exploit higher MAC (multiply-and-accumulation) parallelism that is traditionally uneconomic in CMOS but now efficient in RRAM crossbar. In addition, it can work in a pipelined manner with high computing throughput. Experiment results show that the RRAM-CORDIC algorithm can sustain high accuracy on different transcendental functions, and deliver less than 0.5% NN accuracy loss on typical DNN inference. The elimination of CMOS circuit in turn can trade more computing resources for MVM in the same area budget that improves the performance up to 47% for different networks.

Coarse-grained reconfigurable array (CGRA) driven by dataflow execution is gaining reviving interest as an accelerator architecture of higher energy efficiency. However, with wider adoption in a variety of applications, it is facing complex data and control flows that cause multi-rate execution across different dataflow graphs in CGRAs which degrades the performance. In this paper, we propose a unified storage structure to decouple the multi-rate dataflow for decoupled execution. The structure leverages small distributed buffers with lightweight control. By chaining or aligning these buffers to form larger storage with different control schemes, it caters for different needs of dataflow decoupling when kernels are mapped onto CGRAs. Our experiment results show that by applying the proposed structure in conventional CGRAs, it can save dozens of PEs for dataflow computing, and improve the CGRA performance by an average of 2.53× for applications from different domains. Therefore, we provide a more efficient CGRA design when facing multi-rate dataflow execution.

Digital predistortion (DPD) has played an important role in current forth-generation (4G) mobile system to enhance the linearity of power amplifier (PA) and is expected to continue its functionality in forth-coming 5G. However, large signal bandwidth in 5G scenario complicates the implementation of DPD, especially the feedback path which DPD employs to observe PA output. In this paper, a novel feedback strategy is proposed for DPD of future wideband power amplifier (PA), using ADC with low sampling rate and narrow analog bandwidth. Some issues on practical hardware implementation are addressed to strengthen the realizability of our proposed approach. Simulation results show that similar linearization performance of PA can be achieved compared to conventional DPD, with much lower feedback cost.

Digital predistortion (DPD) technique has already been studied and applied in traditional SISO system. However, when it comes to multiple-input multiple- output (MIMO) systems, new problems such as crosstalk effects must be taken into consideration. In this paper, we propose a neural network based DPD scheme assisted by iterative learning control (ILC) for MIMO system. The proposed method uses direct learning architecture in DPD and can combat nonlinear crosstalk effects. Simulation results show that the proposed method achieves similar or even better performance than the existing polynomial based approach in the presence of nonlinear crosstalk. In addition, the proposed method is computational efficient and is easier for implementation.

Baseband predistortion is a popular and efficient method to linearize high power amplifier (HPA) in wireless communication systems. Polynomial (POLY) and look-up-table (LUT) are two methods to design baseband predistorter (PD). However, on the one hand, POLY-based method is complex to implement. On the other hand, LUT-based predistorter suffers convergence time and quantization error problem. In this paper, we propose a hybrid POLY and LUT predistorter for memory nonlinear system in wideband scenarios, it is also suitable for memoryless channel. Simulations show that the proposed hybrid structure outperforms the traditional one with lower complexity.