Anurag Kumar
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Anurag Kumar has a total of 54 co-authors. Here are the top co-authors by number of shared publications:
- Joy Kuri
ECE Department and the ESE Department, Indian Institute of Science Bangalore (IISc), Bengaluru, India
4 shared publications - Arpan Chattopadhyay
Department of Electrical Engineering, Indian Institute of Technology Delhi, Hauz Khas, New Delhi, India
3 shared publications - Buye Xu
Meta Reality Labs Research
3 shared publications - Shivam Vinayak Vatsa
ECE Department and the ESE Department, Indian Institute of Science Bangalore (IISc), Bengaluru, India
2 shared publications Vamsi Krishna Ithapu
2 shared publications- Aashish Kumar Bohre
Department of Electrical Engineering, National Institute of Technology Durgapur, Durgapur, India
1 shared publication
Latest Publications
- 25 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Signal Processing
- Vol. 74 (1053587X) , 2114-2127
Quickest Change Detection for UAV-Based Sensing
This paper addresses the problem of quickest change detection (QCD) at two spatially separated locations monitored by a single unmanned aerial vehicle (UAV) equipped with a sensor. At any location, the UAV observes i.i.d. data sequentially in discrete time instants. The distribution of the observation data changes at some unknown, arbitrary time and the UAV has to detect this change in the shortest possible time. Change can occur at most at one location over the entire infinite time horizon. The UAV switches between these two locations in order to quickly detect the change. To this end, we propose Location Switching and Change Detection (LS-CD) algorithm which uses a repeated one-sided sequential probability ratio test (SPRT) based mechanism for observation-driven location switching and change detection. The primary goal is to minimize the worst-case average detection delay (WADD) while meeting constraints on the average run length to false alarm (ARL2FA) and the UAV’s time-averaged energy consumption. We provide a rigorous theoretical analysis of the algorithm’s performance by using theory of random walk. Specifically, we derive tight upper and lower bounds to its ARL2FA and a tight upper bound to its WADD. In the special case of a symmetrical setting, our analysis leads to a new asymptotic upper bound to the ARL2FA of the standard CUSUM algorithm, a novel contribution not available in the literature, to our knowledge. Numerical simulations demonstrate the efficacy of LS-CD.
- 25 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Signal Processing
- Vol. 74 (1053587X) , 2114-2127
Quickest Change Detection for UAV-Based Sensing
This paper addresses the problem of quickest change detection (QCD) at two spatially separated locations monitored by a single unmanned aerial vehicle (UAV) equipped with a sensor. At any location, the UAV observes i.i.d. data sequentially in discrete time instants. The distribution of the observation data changes at some unknown, arbitrary time and the UAV has to detect this change in the shortest possible time. Change can occur at most at one location over the entire infinite time horizon. The UAV switches between these two locations in order to quickly detect the change. To this end, we propose Location Switching and Change Detection (LS-CD) algorithm which uses a repeated one-sided sequential probability ratio test (SPRT) based mechanism for observation-driven location switching and change detection. The primary goal is to minimize the worst-case average detection delay (WADD) while meeting constraints on the average run length to false alarm (ARL2FA) and the UAV’s time-averaged energy consumption. We provide a rigorous theoretical analysis of the algorithm’s performance by using theory of random walk. Specifically, we derive tight upper and lower bounds to its ARL2FA and a tight upper bound to its WADD. In the special case of a symmetrical setting, our analysis leads to a new asymptotic upper bound to the ARL2FA of the standard CUSUM algorithm, a novel contribution not available in the literature, to our knowledge. Numerical simulations demonstrate the efficacy of LS-CD.
- 28 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Networking
- Vol. 34, 989-1004
Optimum MultiAP Coordination in Wi-Fi With Overlay Time-Sliced Scheduling
We report on the design and implementation of multiAP (Access Point) coordination in a Wi-Fi network, using fine-grained overlay time-slicing, centralized queueing and link activity scheduling. Our approach achieves network wide utility optimal downlink and uplink TCP throughputs, while permitting rate guarantees to designated TCP flows. We also obtain significant improvement in the performance of downlink streaming video, in the presence of cochannel interference from other APs. This work is a substantial extension of the ADWISER (ADvanced WI-fi Service EnhanceR) system reported in Hegde et al. (2013), Sunny et al. (2017), and Sevani et al. (2022). ADWISER is a central scheduler through which all downlink and uplink traffic passes; it overlays periodic time-slices, during which sets of AP-STA links are scheduled. During the time-slices the scheduled AP-STA links use the default IEEE 802.11 PHY/MAC mechanisms; thus ADWISER works without any modifications to the APs and the STAs. Our earlier work required the knowledge of the sets of the AP-STA links that can be scheduled without mutual interference between them, and used time-slices of 100s of milliseconds, which were statically partitioned into scheduling slots to which the known sets of AP-STA links were mapped. The techniques that we now develop and implement perform fine-grained time-slicing, down to 20 ms, thus permitting the handling of delay sensitive Internet applications. ADWISER, now, only needs to know the AP-STA associations. By a process of scheduling and measuring throughputs, using stochastic approximation algorithms, ADWISER dynamically determines the set of AP-STA links to schedule in successive time-slices, thereby achieving global network utility optimal throughputs. We have implemented ADWISER on a Linux OS based server machine, and all performance results we report are from test-beds that use commercial off-the-shelf APs and laptops as STAs.
- 28 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Networking
- Vol. 34, 989-1004
Optimum MultiAP Coordination in Wi-Fi With Overlay Time-Sliced Scheduling
We report on the design and implementation of multiAP (Access Point) coordination in a Wi-Fi network, using fine-grained overlay time-slicing, centralized queueing and link activity scheduling. Our approach achieves network wide utility optimal downlink and uplink TCP throughputs, while permitting rate guarantees to designated TCP flows. We also obtain significant improvement in the performance of downlink streaming video, in the presence of cochannel interference from other APs. This work is a substantial extension of the ADWISER (ADvanced WI-fi Service EnhanceR) system reported in Hegde et al. (2013), Sunny et al. (2017), and Sevani et al. (2022). ADWISER is a central scheduler through which all downlink and uplink traffic passes; it overlays periodic time-slices, during which sets of AP-STA links are scheduled. During the time-slices the scheduled AP-STA links use the default IEEE 802.11 PHY/MAC mechanisms; thus ADWISER works without any modifications to the APs and the STAs. Our earlier work required the knowledge of the sets of the AP-STA links that can be scheduled without mutual interference between them, and used time-slices of 100s of milliseconds, which were statically partitioned into scheduling slots to which the known sets of AP-STA links were mapped. The techniques that we now develop and implement perform fine-grained time-slicing, down to 20 ms, thus permitting the handling of delay sensitive Internet applications. ADWISER, now, only needs to know the AP-STA associations. By a process of scheduling and measuring throughputs, using stochastic approximation algorithms, ADWISER dynamically determines the set of AP-STA links to schedule in successive time-slices, thereby achieving global network utility optimal throughputs. We have implemented ADWISER on a Linux OS based server machine, and all performance results we report are from test-beds that use commercial off-the-shelf APs and laptops as STAs.
- 3 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Control of Network Systems
- Vol. 12 (1) , 749-762
A Low-Delay MAC for IoT Applications: Decentralized Optimal Scheduling of Queues Without Explicit State Information Sharing
For a system of collocated nodes sharing a time-slotted wireless channel, we seek a medium access control that provides low mean delay, has distributed control, and does not require explicit exchange of state information or control signals. We consider a practical information structure where each node has local information and some common information obtained from overhearing. We approach the problem via two steps: 1) we show that it is sufficient for the policy to be “greedy” and “exhaustive”; limiting the policy to this class reduces the problem to obtaining a queue switching policy at queue emptiness instants; and 2) by formulating the delay optimal scheduling as a partially observed Markov decision process, we show that the optimal switching rule is stochastic largest queue. Using this theory as the basis, we develop a practical, tunable, distributed scheduler, QZMAC, which is an extension to the existing ZMAC protocol. We implement QZMAC on standard off-the-shelf TelosB motes and also use simulations to compare QZMAC with the full-knowledge centralized scheduler and with ZMAC. We use our implementation to study the impact of false detection, while overhearing the common information, and the efficiency of QZMAC. Simulation results show that the mean delay with QZMAC is close to that of the full-knowledge centralized scheduler.
- 3 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Control of Network Systems
- Vol. 12 (1) , 749-762
A Low-Delay MAC for IoT Applications: Decentralized Optimal Scheduling of Queues Without Explicit State Information Sharing
For a system of collocated nodes sharing a time-slotted wireless channel, we seek a medium access control that provides low mean delay, has distributed control, and does not require explicit exchange of state information or control signals. We consider a practical information structure where each node has local information and some common information obtained from overhearing. We approach the problem via two steps: 1) we show that it is sufficient for the policy to be “greedy” and “exhaustive”; limiting the policy to this class reduces the problem to obtaining a queue switching policy at queue emptiness instants; and 2) by formulating the delay optimal scheduling as a partially observed Markov decision process, we show that the optimal switching rule is stochastic largest queue. Using this theory as the basis, we develop a practical, tunable, distributed scheduler, QZMAC, which is an extension to the existing ZMAC protocol. We implement QZMAC on standard off-the-shelf TelosB motes and also use simulations to compare QZMAC with the full-knowledge centralized scheduler and with ZMAC. We use our implementation to study the impact of false detection, while overhearing the common information, and the efficiency of QZMAC. Simulation results show that the mean delay with QZMAC is close to that of the full-knowledge centralized scheduler.
- 13 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 1, 4843-4848
Quickest Change Point Detection with Measurements over a Lossy Link
Motivated by Industry 4.0 applications, we consider quickest change point detection (QCD) when process measurements are transmitted by a sensor over a lossy wireless link to a decision maker (DM). The sensor node samples measurements using a Bernoulli sampling process, and places the measurement samples in a transmit queue of the transmitter. The transmitter uses a retransmit-until-success transmission strategy to deliver packets to the DM over the lossy link, which is modeled as an independent Bernoulli process and has different loss probabilities before and after the change. We pose the QCD problem in the non-Bayesian setting under Lorden's framework [1], and derive a CUSUM algorithm. By defining a suitable Markov process, involving the DM measurements and the queue length process, we show that the problem reduces to QCD of a Markov process. Characterizing the information measure $I$ per measurement sample at the DM, our analysis proves the asymptotic optimality of our algorithm when the false alarm rate tends to zero. We discuss extensions of the analysis to periodic sampling and no-retransmission cases. Through numerical analysis, we demonstrate trade-offs that can be used to optimize system design parameters such as the sampling rate of the measurement process in the non-asymptotic regime.- 13 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 1, 4843-4848
Quickest Change Point Detection with Measurements over a Lossy Link
Motivated by Industry 4.0 applications, we consider quickest change point detection (QCD) when process measurements are transmitted by a sensor over a lossy wireless link to a decision maker (DM). The sensor node samples measurements using a Bernoulli sampling process, and places the measurement samples in a transmit queue of the transmitter. The transmitter uses a retransmit-until-success transmission strategy to deliver packets to the DM over the lossy link, which is modeled as an independent Bernoulli process and has different loss probabilities before and after the change. We pose the QCD problem in the non-Bayesian setting under Lorden's framework [1], and derive a CUSUM algorithm. By defining a suitable Markov process, involving the DM measurements and the queue length process, we show that the problem reduces to QCD of a Markov process. Characterizing the information measure $I$ per measurement sample at the DM, our analysis proves the asymptotic optimality of our algorithm when the false alarm rate tends to zero. We discuss extensions of the analysis to periodic sampling and no-retransmission cases. Through numerical analysis, we demonstrate trade-offs that can be used to optimize system design parameters such as the sampling rate of the measurement process in the non-asymptotic regime.- 15 September 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 982-988
Reliability Based Performance Analysis of Distribution Network with Dispersed Generation Using Optimization
By distributing renewable energy at the distribution level, DG integration can assist in achieving both economic and environmental goals. The primary objective is to improve system's reliability also lower energy expenses by providing backup power during outages. Therefore, optimizing DG allocation can lead to a more effective, sustainable, and cost-efficient distribution system. DG has gained a significant presence in power networks worldwide, and it continues growing in developing and developed nations. One of the main goals of DG planning is loss reduction, Improved voltage profile, more excellent stability, and reduced power losses; DG installation should be carried out sequentially. This paper describes where the DG should be placed in the IEEE 69 systems and IEEE 33 bus test systems. To improve the reliability of power distribution networks, the project aims to increase voltage stability and decrease power losses. The distribution systems are modeled using the ETAP and MATLAB by Particle Swarm Optimization (PSO) technique. Three alternative scenarios—without DG, with one DG, and with two DG, are considered to demonstrate the improvement brought about by DG deployment.
- 15 September 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 982-988
Reliability Based Performance Analysis of Distribution Network with Dispersed Generation Using Optimization
By distributing renewable energy at the distribution level, DG integration can assist in achieving both economic and environmental goals. The primary objective is to improve system's reliability also lower energy expenses by providing backup power during outages. Therefore, optimizing DG allocation can lead to a more effective, sustainable, and cost-efficient distribution system. DG has gained a significant presence in power networks worldwide, and it continues growing in developing and developed nations. One of the main goals of DG planning is loss reduction, Improved voltage profile, more excellent stability, and reduced power losses; DG installation should be carried out sequentially. This paper describes where the DG should be placed in the IEEE 69 systems and IEEE 33 bus test systems. To improve the reliability of power distribution networks, the project aims to increase voltage stability and decrease power losses. The distribution systems are modeled using the ETAP and MATLAB by Particle Swarm Optimization (PSO) technique. Three alternative scenarios—without DG, with one DG, and with two DG, are considered to demonstrate the improvement brought about by DG deployment.
- Open Access
LA-VocE: Low-SNR Audio-visual Speech Enhancement using Neural Vocoders
Audio-visual speech enhancement aims to extract clean speech from a noisy environment by leveraging not only the audio itself but also the target speaker's lip movements. This approach has been shown to yield improvements over audio-only speech enhancement, particularly for the removal of interfering speech. Despite recent advances in speech synthesis, most audio-visual approaches continue to use spectral mapping/masking to reproduce the clean audio, often resulting in visual backbones added to existing speech enhancement architectures. In this work, we propose LA-VocE, a new two-stage approach that predicts mel-spectrograms from noisy audio-visual speech via a transformer-based architecture, and then converts them into waveform audio using a neural vocoder (HiFi-GAN). We train and evaluate our framework on thousands of speakers and 11+ different languages, and study our model's ability to adapt to different levels of background noise and speech interference. Our experiments show that LA-VocE outperforms existing methods according to multiple metrics, particularly under very noisy scenarios.
Open AccessLA-VocE: Low-SNR Audio-visual Speech Enhancement using Neural Vocoders
Audio-visual speech enhancement aims to extract clean speech from a noisy environment by leveraging not only the audio itself but also the target speaker's lip movements. This approach has been shown to yield improvements over audio-only speech enhancement, particularly for the removal of interfering speech. Despite recent advances in speech synthesis, most audio-visual approaches continue to use spectral mapping/masking to reproduce the clean audio, often resulting in visual backbones added to existing speech enhancement architectures. In this work, we propose LA-VocE, a new two-stage approach that predicts mel-spectrograms from noisy audio-visual speech via a transformer-based architecture, and then converts them into waveform audio using a neural vocoder (HiFi-GAN). We train and evaluate our framework on thousands of speakers and 11+ different languages, and study our model's ability to adapt to different levels of background noise and speech interference. Our experiments show that LA-VocE outperforms existing methods according to multiple metrics, particularly under very noisy scenarios.
- Open Access
Leveraging Heteroscedastic Uncertainty in Learning Complex Spectral Mapping for Single-channel Speech Enhancement
Most speech enhancement (SE) models learn a point estimate and do not make use of uncertainty estimation in the learning process. In this paper, we show that modeling heteroscedastic uncertainty by minimizing a multivariate Gaussian negative log-likelihood (NLL) improves SE performance at no extra cost. During training, our approach augments a model learning complex spectral mapping with a temporary submodel to predict the covariance of the enhancement error at each time-frequency bin. Due to unrestricted heteroscedastic uncertainty, the covariance introduces an undersampling effect, detrimental to SE performance. To mitigate undersampling, our approach inflates the uncertainty lower bound and weights each loss component with their uncertainty, effectively compensating severely undersampled components with more penalties. Our multivariate setting reveals common covariance assumptions such as scalar and diagonal matrices. By weakening these assumptions, we show that the NLL achieves superior performance compared to popular loss functions including the mean squared error (MSE), mean absolute error (MAE), and scale-invariant signal-to-distortion ratio (SI-SDR).
Open AccessLeveraging Heteroscedastic Uncertainty in Learning Complex Spectral Mapping for Single-channel Speech Enhancement
Most speech enhancement (SE) models learn a point estimate and do not make use of uncertainty estimation in the learning process. In this paper, we show that modeling heteroscedastic uncertainty by minimizing a multivariate Gaussian negative log-likelihood (NLL) improves SE performance at no extra cost. During training, our approach augments a model learning complex spectral mapping with a temporary submodel to predict the covariance of the enhancement error at each time-frequency bin. Due to unrestricted heteroscedastic uncertainty, the covariance introduces an undersampling effect, detrimental to SE performance. To mitigate undersampling, our approach inflates the uncertainty lower bound and weights each loss component with their uncertainty, effectively compensating severely undersampled components with more penalties. Our multivariate setting reveals common covariance assumptions such as scalar and diagonal matrices. By weakening these assumptions, we show that the NLL achieves superior performance compared to popular loss functions including the mean squared error (MSE), mean absolute error (MAE), and scale-invariant signal-to-distortion ratio (SI-SDR).
- Open Access
Improving Speech Enhancement through Fine-Grained Speech Characteristics
While deep learning based speech enhancement systems have made rapid progress in improving the quality of speech signals, they can still produce outputs that contain artifacts and can sound unnatural. We propose a novel approach to speech enhancement aimed at improving perceptual quality and naturalness of enhanced signals by optimizing for key characteristics of speech. We first identify key acoustic parameters that have been found to correlate well with voice quality (e.g. jitter, shimmer, and spectral flux) and then propose objective functions which are aimed at reducing the difference between clean speech and enhanced speech with respect to these features. The full set of acoustic features is the extended Geneva Acoustic Parameter Set (eGeMAPS), which includes 25 different attributes associated with perception of speech. Given the non-differentiable nature of these feature computation, we first build differentiable estimators of the eGeMAPS and then use them to fine-tune existing speech enhancement systems. Our approach is generic and can be applied to any existing deep learning based enhancement systems to further improve the enhanced speech signals. Experimental results conducted on the Deep Noise Suppression (DNS) Challenge dataset shows that our approach can improve the state-of-the-art deep learning based enhancement systems.
Open AccessImproving Speech Enhancement through Fine-Grained Speech Characteristics
While deep learning based speech enhancement systems have made rapid progress in improving the quality of speech signals, they can still produce outputs that contain artifacts and can sound unnatural. We propose a novel approach to speech enhancement aimed at improving perceptual quality and naturalness of enhanced signals by optimizing for key characteristics of speech. We first identify key acoustic parameters that have been found to correlate well with voice quality (e.g. jitter, shimmer, and spectral flux) and then propose objective functions which are aimed at reducing the difference between clean speech and enhanced speech with respect to these features. The full set of acoustic features is the extended Geneva Acoustic Parameter Set (eGeMAPS), which includes 25 different attributes associated with perception of speech. Given the non-differentiable nature of these feature computation, we first build differentiable estimators of the eGeMAPS and then use them to fine-tune existing speech enhancement systems. Our approach is generic and can be applied to any existing deep learning based enhancement systems to further improve the enhanced speech signals. Experimental results conducted on the Deep Noise Suppression (DNS) Challenge dataset shows that our approach can improve the state-of-the-art deep learning based enhancement systems.
- Open Access
SAQAM: Spatial Audio Quality Assessment Metric
Audio quality assessment is critical for assessing the perceptual realism of sounds. However, the time and expense of obtaining ''gold standard'' human judgments limit the availability of such data. For AR&VR, good perceived sound quality and localizability of sources are among the key elements to ensure complete immersion of the user. Our work introduces SAQAM which uses a multi-task learning framework to assess listening quality (LQ) and spatialization quality (SQ) between any given pair of binaural signals without using any subjective data. We model LQ by training on a simulated dataset of triplet human judgments, and SQ by utilizing activation-level distances from networks trained for direction of arrival (DOA) estimation. We show that SAQAM correlates well with human responses across four diverse datasets. Since it is a deep network, the metric is differentiable, making it suitable as a loss function for other tasks. For example, simply replacing an existing loss with our metric yields improvement in a speech-enhancement network.
Open AccessSAQAM: Spatial Audio Quality Assessment Metric
Audio quality assessment is critical for assessing the perceptual realism of sounds. However, the time and expense of obtaining ''gold standard'' human judgments limit the availability of such data. For AR&VR, good perceived sound quality and localizability of sources are among the key elements to ensure complete immersion of the user. Our work introduces SAQAM which uses a multi-task learning framework to assess listening quality (LQ) and spatialization quality (SQ) between any given pair of binaural signals without using any subjective data. We model LQ by training on a simulated dataset of triplet human judgments, and SQ by utilizing activation-level distances from networks trained for direction of arrival (DOA) estimation. We show that SAQAM correlates well with human responses across four diverse datasets. Since it is a deep network, the metric is differentiable, making it suitable as a loss function for other tasks. For example, simply replacing an existing loss with our metric yields improvement in a speech-enhancement network.
- Open Access
- 15 April 2020
- Published by Technoscience Academy in International Journal of Scientific Research in Computer Science, Engineering and Information Technology
- p. 490-494
Improvement in K-medoid Clustering using Density based Node Selection
The wireless sensor networks have become a very hot topic of research in the recent years because of their wide range of applications such as industrial and agricultural monitoring, military surveillance, smart homes etc. These sensors can also be used in locations that are potentially hazardous to the human beings or are just out of our reach at this moment of time. A wireless sensor network is collection of large number of individual sensor nodes. A sensor usually comprises of components for sensing the environment, processing, communicating and a power supply. The major limiting factor of the wireless sensor networks is the limited amount of power that each of the sensors can carry and also the energy in these sensors cannot be replenished easily. So, the major design challenge in a wireless sensor network has always been about reduction in the consumption of energy by the sensors. Clustering is one of the popular methods to reduce energy consumption in wireless sensor networks. Here, we propose a scheme to decrease the energy consumption and prolong the lifetime of wireless sensor network. The main idea behind the scheme is that we try to minimise the communication distance between the individual sensor nodes using the clustering technique.
- 15 April 2020
- Published by Technoscience Academy in International Journal of Scientific Research in Computer Science, Engineering and Information Technology
- p. 490-494
Open AccessImprovement in K-medoid Clustering using Density based Node Selection
The wireless sensor networks have become a very hot topic of research in the recent years because of their wide range of applications such as industrial and agricultural monitoring, military surveillance, smart homes etc. These sensors can also be used in locations that are potentially hazardous to the human beings or are just out of our reach at this moment of time. A wireless sensor network is collection of large number of individual sensor nodes. A sensor usually comprises of components for sensing the environment, processing, communicating and a power supply. The major limiting factor of the wireless sensor networks is the limited amount of power that each of the sensors can carry and also the energy in these sensors cannot be replenished easily. So, the major design challenge in a wireless sensor network has always been about reduction in the consumption of energy by the sensors. Clustering is one of the popular methods to reduce energy consumption in wireless sensor networks. Here, we propose a scheme to decrease the energy consumption and prolong the lifetime of wireless sensor network. The main idea behind the scheme is that we try to minimise the communication distance between the individual sensor nodes using the clustering technique.