Estimating the height of buildings only based on monocular satellite images is a difficult task due to limited annotated data quantity, subpar data quality, changes in building kinds, and varying angles of satellite perspective, light and shadow. Given the task of predicting a single height value for each building in a satellite image, this work proposes a method for instance segmentation based building height estimation. In a single end to end multitask model, the system predicts contours and corresponding heights of buildings from a monocular remote sensing image. Using a dataset curated for this study, we present initial results on building height classification with three classes. Additionally we use transfer learning from a public building contour dataset to improve results. Overall, an average classification accuracy of 70% is achieved for buildings in each standalone height class.
In this study, SPICE simulation of the CISPR 25 test setup has been carried out to observe the effect of different structures on the electronic board designed for the LED driver of side marker lamp of a car on the conducted emission (CE) level. Therefore, four different SPICE models have been created. In addition to these, the CE test for this electronic board has been carried out in a semi-anechoic chamber. Comparing simulation results with the actual test results, the closest results have been found in the analysis of the fourth model. In the frequency range of 150 kHz - 30 MHz, the highest CE values have been found as 58.13 dBnV at 150 kHz in the fourth model (added vias), while it has been 90.17 dBnV at 213.13kHz in the actual test results. Results for 30 MHz - 108 MHz have been found as 71.47 dBnV at 105 MHz and 88.82 dBnV at 59.39 MHz, respectively.
We show how to address nonlinearities in power amplifiers (PAs), which limit the power efficiency of mobile devices, increase the error vector magnitude, using an deep neural-network (DNN) method. DPD is frequently performed using polynomial-based algorithms that employ an indirect-learning architecture (ILA), which can be computationally complex, particularly on mobile devices, and highly sensitive to noise. By first training a DNN to model the PA and then training a predistorter using PA data through the PA DNN model. The DNN DPD successfully learns the unique PA distortions that a polynomial-based model may struggle to fit, and therefore may provide a nice balance between computation cost and DPD efficiency. We use two different DNN models to show the performance of our DNN approach and examine the complexity tradeoffs.
Adaptive Beamforming is an array signal processing problem in which a beam pattern is generated in the direction of desired signal and nulls are placed in the directions of undesired signals. Neural networks are widely employed for adaptive beamforming problems. In this paper, a autoencoder(AE) guided radial basis function(RBF) neural network(NN) is proposed. The proposed AE guided RBF neural network is also named as hybrid neural network since it is a combination of AE and RBF. Feature extraction and data compression are achieved by using the encoder part of AE in the input and first hidden layer of the hybrid neural network. The proposed hybrid neural network model is compared with a classical 3-layered RBF neural network according to Signal to Interferance Ratio(SIR) performance, Half Power Beamwidth(HPBW) and processing load. Simulation results show that the proposed hybrid neural network reduces the processing load without decreasing the SIR performance.
In recent years, quantum computer development projects have progressed considerably in terms of quality and quantity. These studies also pose a serious risk for widely used security systems based on classical public-key cryptography algorithms. Therefore, development of hardware and software applications that use post-quantum methods are considered a necessity both for today and for the future. Kyber is one of the most promising key encapsulation mechanism in the final round of NIST post-quantum cryptography standardization process. In this study, we present a 64-bit application specific transport triggered architecture processor which has custom operations for accelerating the Kyber algorithm. Moreover, we compare the processor and its various versions with a 64-bit RISC-V core in terms of performance, total energy consumption and resource utilization. According to the test results, our custom processor provides at least 2x better performance and 1.5x lower energy consumption while increasing the required chip area approximately 1.4x.
