Payment apps and digital wallets are powerful tools used to exchange e-money via the internet. However, with the progressive disappearance of cash, there is a need for the digital equivalent of physical banknotes to guarantee the same level of anonymity of private payments. Few efforts to solve the double-spending problem exist in P2P payments (i.e., in avoiding the possibility of a payer retaining copies of digital coins in absence of a trusted third party (TTP)), and further research efforts are needed to explore options to preserve the privacy of payments, as per the mandates of numerous central bank digital currency (CBDC) exploratory initiatives, such as the digital euro. Moreover, generic programmability requirements and energetic impacts should be considered. In this paper, we present a sustainable offline P2P payment scheme to face the double-spending problem by means of a one-time program (OTP) approach. The approach consists of wiping the business logic out of a client’s app and allowing financial intermediaries to inject a certified payment code into the user’s device, which will execute (asynchronously and offline) at the time of payment. To do so, we wrap each coin in a program at the time of withdrawal. Then the program exploits the trusted execution environment (TEE) of modern smartphones to transfer itself from the payer to the payee via a direct IoT link. To confirm the validity of the approach, we performed qualitative and quantitative evaluations, specifically focusing on the energetic sustainability of the proposed scheme. Results show that our payment scheme is energetically sustainable as the current absorbed for sending one coin is, at most, ~1.8 mAh on an Apple smartphone. We advance the state-of-the-art because the scheme meets the programmability, anonymity, and sustainability requirements (at the same time).

The Internet of Things (IoT) is set to play a significant role in the future development of smart cities, which are designed to be environmentally friendly. However, the proliferation of these devices, along with their frequent replacements and the energy required to power them, contributes to a significant environmental footprint. In this paper we provide scientific evidences on the advantages of using an energy design pattern named ‘Batch Operations’ (BO) to optimize energy consumption on mobile devices. Big ICT companies like Google already batch multiple API calls instead of putting the device into an active state many times. This is supposed to save tail energy consumption in communication peripherals. To confirm this, we set up an experiment where we compare energy consumption and carbon emission when BO is applied to two communication peripherals on Android mobile device: 4G and GPS. Results show that (1) BO can save up to 40% energy when sending HTTP requests, resulting in an equivalent reduction in $C{O}_2$ emissions. (2) no advantages for the GPS interface.

The design of detectors with constant false alarm rate (CFAR) property is a cornerstone challenge in radar signal processing, for both Gaussian and non-Gaussian environments. To this aim, a forward-thinking strategy is to construct decision statistics as functions of suitable maximal invariants associated with invariant tests that guarantee, through a suitable transformation group, the CFAR property by construction. However, the distribution of such maximal invariants is often difficult to handle, especially for complicated non-Gaussian models, making the derivation of the generalized likelihood ratio test (GLRT) challenging. In this work, we introduce a novel learning-based CFAR detection framework, in which a trained probabilistic encoder maps maximal invariant statistics (or functions thereof) to a convenient low-dimensional latent space where a latent GLRT-based detector (L-GLRT) is easy to derive. More specifically, a cross-entropy loss function with Kullback-Leibler (KL) divergence regularization is adopted to encourage the latent distributions under both $H_{0}$ (target free) and $H_{1}$ (target present) hypotheses to be as close as possible, in an information-theoretic sense, to Gaussian densities. Mismatched data incorporated under either hypotheses are introduced to promote robustness or selectivity. The approach unifies Gaussian and non-Gaussian settings, spanning from point- like to extended targets under the complex multivariate elliptically contoured matrix (CMECM) family, and is benchmarked against state-of-the-art classical and data-driven detectors. Moreover, since the latent space low dimensional, insightful visualization of the behavior of the designed L-GLRT detectors can be obtained. Numerical results show that the proposed method achieves superior robustness/selectivity trade-offs while preserving CFAR guarantees by design and containing $P_{d}$ losses under matched conditions.

Data Assimilation (DA) is a type of data fusion approach that integrates observational data with the simulated state produced by a numerical model. One of the primary objectives of DA is to enhance the accuracy of numerical simulations by achieving a more accurate and realistic representation of the system’s state. This method is extensively applied in fields such as weather forecasting and oceanography, where it plays a crucial role in improving model accuracy. In these disciplines, two of the most commonly employed DA approaches are the variational methods and sequential methods such as the Kalman Filter and the Kalman Smoother. Although effective, all such methods are computationally expensive. Recently, deep learning algorithms have been developed to reduce the time required by classical assimilation methods while maintaining or improving assimilation accuracy. We propose a novel architectural framework based on a Conditional Denoising Diffusion Probabilistic Model (CDDPM), incorporating a land-masking strategy and a novel latent-space interpolation mechanism. While the approach is designed to be domain-independent for geophysical data assimilation, we demonstrate its effectiveness in performing simultaneous assimilation and bias correction over the complex Mediterranean Sea domain. In addition, to improve CDDPM assimilation skills, we also fine-tuned our model by adding a tail convolutional network that conforms CDDPM output with satellite observations. Furthermore, by training on ocean simulation-reanalysis pairs, the CDDPM learns how to generate corrected ocean state variables that are statistically closer to the reference fields, even without observational input at inference time. This dual capability allows the model to operate flexibly under different data availability scenarios. Extensive experiments show that our data-driven model achieves good performance similar to, and in some cases better than the reference system, demonstrating the ability to generate a state closer to satellite observations while maintaining consistency with numerical solutions.

Data Assimilation (DA) is a type of data fusion approach that integrates observational data with the simulated state produced by a numerical model. One of the primary objectives of DA is to enhance the accuracy of numerical simulations by achieving a more accurate and realistic representation of the system’s state. This method is extensively applied in fields such as weather forecasting and oceanography, where it plays a crucial role in improving model accuracy. In these disciplines, two of the most commonly employed DA approaches are the variational methods and sequential methods such as the Kalman Filter and the Kalman Smoother. Although effective, all such methods are computationally expensive. Recently, deep learning algorithms have been developed to reduce the time required by classical assimilation methods while maintaining or improving assimilation accuracy. We propose a novel architectural framework based on a Conditional Denoising Diffusion Probabilistic Model (CDDPM), incorporating a land-masking strategy and a novel latent-space interpolation mechanism. While the approach is designed to be domain-independent for geophysical data assimilation, we demonstrate its effectiveness in performing simultaneous assimilation and bias correction over the complex Mediterranean Sea domain. In addition, to improve CDDPM assimilation skills, we also fine-tuned our model by adding a tail convolutional network that conforms CDDPM output with satellite observations. Furthermore, by training on ocean simulation-reanalysis pairs, the CDDPM learns how to generate corrected ocean state variables that are statistically closer to the reference fields, even without observational input at inference time. This dual capability allows the model to operate flexibly under different data availability scenarios. Extensive experiments show that our data-driven model achieves good performance similar to, and in some cases better than the reference system, demonstrating the ability to generate a state closer to satellite observations while maintaining consistency with numerical solutions.

Recent work has shown that detection of AI-generated images can be performed based on low-level fingerprints extracted from the frequency spectrum of noise residuals, the latter obtained by subtracting the high-level image content through a denoising filter. However, state-of-the-art detectors are based on deep neural networks with tens of millions of parameters, which also require large datasets for training. In this letter, we show that comparable detection performance can be achieved at a fraction of memory and computational requirements, by using a sparse spectral representation coupled with a lightweight classifier. Specifically, we propose a simple yet effective sparsification algorithm able to adaptively extracts the dominant spectral peaks, which reduces the amount of data to be stored and processed to only 0.14% on average, speeding up in turn both the training and execution of the detection algorithm. We compare the proposed approach against state-of-the-art deep neural networks, including ablation studies to highlight complexity-performance trade-offs when removing sparsification (so processing plain noise residual spectra) or even noise residual spectra extraction altogether (so processing pixel-domain data). Further reduction in complexity is achieved by adopting a lightweight neural network, which lowers the number of parameters by about two orders of magnitude compared to deep detectors. Alternative adoption of support vector machine is also investigated. Results show that it is possible to attain accuracy close to state-of-the-art deep neural networks with significantly lower complexity and smaller training datasets.