The greatest technological transformation the world has ever seen was brought about by artificial intelligence (AI). It presents significant opportunities for the financial sector to enhance risk management, democratize financial services, ensure consumer protection, and improve customer experience. Modern machine learning models are more accessible than ever, but it has been challenging to create and implement systems that support real-world financial applications, primarily due to their lack of transparency and explainability—both of which are essential for building trustworthy technology. The novelty of this study lies in the development of an explainable AI (XAI) model that not only addresses these transparency concerns but also serves as a tool for policy development in credit risk management. By offering a clear understanding of the underlying factors influencing AI predictions, the proposed model can assist regulators and financial institutions in shaping data-driven policies, ensuring fairness, and enhancing trust. This study proposes an explainable AI model for credit risk management, specifically aimed at quantifying the risks associated with credit borrowing through peer-to-peer lending platforms. The model leverages Shapley values to generate AI predictions based on key explanatory variables. The decision tree and random forest models achieved the highest accuracy levels of 0.89 and 0.93, respectively. The model’s performance was further tested using a larger dataset, where it maintained stable accuracy levels, with the decision tree and random forest models reaching accuracies of 0.90 and 0.93, respectively. To ensure reliable explainable AI (XAI) modeling, these models were chosen due to the binary classification nature of the problem. LIME and SHAP were employed to present the XAI models as both local and global surrogates.
28 September 2018 in IEEE Transactions on Microwave Theory and Techniques
Multi-input multi-output (MIMO) is anticipated to be a prominent technique proposed in the wireless communications to improve the system capacity and data rates of the wireless networks. However, the MIMO transmitter suffers from imperfections, such as crosstalk, power-amplifier (PA) nonlinearity, in-phase and quadrature (I/Q) imbalance, and dc offset. Investigating these effects, this paper proposes neural network (NN)-based digital predistortion (DPD) as an integral solution to compensate for crosstalk, PA nonlinearity, I/Q imbalance, and dc offset imperfections simultaneously in MIMO transmitters. The proposed NN DPD model provides a one-step single-model digital mitigation solution to multibranches of MIMO transmitters. With the increase in the dimensions of MIMO transmitter, the proposed NN-based DPD model provides a better compensation for transmitter imperfections and also reduces the complexity as compared to the state-of-the-art DPD methods. The proof-of-concept is provided with the 2 x 2 and 3 x 3 MIMO transmitters in the presence of strong PA nonlinearity, crosstalk, I/Q imbalance, and dc offset for homodyne as well as heterodyne transmitters' cases.
11 May 2021 in IEEE Transactions on Microwave Theory and Techniques
Neural network (NN)-based models are perceived as being accurate models for power amplifier (PA) behavioral modeling and digital predistortion (DPD) applications. However, the complexity of NN’s models in terms of the number of coefficients increases with the memory depth, nonlinearity order, signal bandwidth, and the number of carriers. In this article, a novel augmented convolutional neural network (ACNN)-based DPD is proposed to linearize the concurrent multiband PAs. In the ACNN model, the convolution layer with a fully connected layer serves to model the intermodulation distortion (IMD), cross-modulation distortion (CMD), and nonlinearities of the concurrent multiband PA. The pooling layer in the ACNN reduces the model’s complexity. The measurement results show that the proposed DPD requires significantly fewer coefficients and floating-point operations (FLOPs) than the state-of-the-art DPDs. This model’s complexity reduction arises with the number of carriers in the multiband aggregated signal. The measurement results also show that the proposed DPD has a better linearization performance in terms of metrics, such as normalized mean square error (NMSE), adjacent channel power ratio (ACPR), and error vector magnitude (EVM).
19 September 2022 in IEEE Transactions on Circuits and Systems II: Express Briefs
In modern wireless compact transmitters, Power Amplifier (PA) behavior is considerably affected by the impedance mismatch between the PA’s output and the antenna’s input. This PA’s output mismatch results in a reflection at the PA–antenna interface. In this brief, reflection-aware PA modeling and digital predistortion (DPD) techniques are proposed to mitigate the negative impact of this mismatch on the forward and reverse models of the PA. An Augmented Convolutional neural network model $(\boldsymbol{\Gamma }$ ACNN) is proposed to linearize a Doherty PA under different values of the output mismatch using a single set of coefficients. The developed DPD shows robust performance metrics like normalized mean square error (NMSE), and adjacent channel power ratio (ACPR) under diverse complex output mismatch levels.
1 October 2018 in IEEE Transactions on Circuits and Systems II: Express Briefs
With the increase in demands of the multi-band and ultra-wideband based transmitter, the research interest in power amplifiers (PAs) is moving towards the designing of multi-band and ultra-wideband PAs. This leads to unprecedented challenges such as the possibility of multiband transmission at harmonic frequencies. This brief investigates and presents the novel DPD models for linearization of a previously uninvestigated case of concurrent tri-band PA at harmonic frequencies. This brief analyzes the intermodulation (IMD) terms which are producing harmonic distortions and proposes three-dimensional harmonic memory polynomial (3D-HMP) and harmonic Volterra spline (3D-HVS) models for mitigation of in-band harmonic, cross-modulation, and intermodulation distortions. The results show that the proposed 3D-HMP and 3D-HVS DPD models have linearization performance in terms of adjacent channel power ratio (ACPR) below -50 dBc.
2 February 2017 in International Journal of RF and Microwave Computer-Aided Engineering