Chapter Four · failure evidence
What Digital Filtering & Signal Denoising got wrong, from 98 dissertations
The records describe various challenges encountered when applying digital filtering and signal denoising methods across engineering and biomedical applications. Key difficulties include signal distortion, phase delay, model misspecification, noise assumption violations, and sophisticated filters failing to surpass simple baselines. These records come from PhD theses at 30 institutions, 2021 to 2026. Each links to its thesis. They were extracted by language models reading the full text, so treat each as a lead to read, not a verdict.
Filtering removes critical signal components and distorts underlying waveform morphology
Applying aggressive bandpass, lowpass, or moving average filters frequently attenuates true signal peaks and erases essential high-frequency dynamics when noise and signal spectra overlap. Researchers rejected or failed with these filters because over-smoothing degraded geometric trajectories, distorted waveform shapes, and discarded informative events.
Tried and failed
Kalman filtering with signal quality indices applied to physiological time series denoising. Outcome: worse than baseline. Reason: Denoising removed motion and noise patterns that were informative and correlated with the target event
Self-Aware Machine Learning for Chronic Pathology Monitoring on Wearable Devices · EPFL
Tried and failed
Kalman filtering with constant process noise covariance applied to physiological sensor time series denoising. Reason: Low process noise oversmoothed dynamic peaks while high process noise failed to suppress high-frequency noise
Tried and failed
classical band-pass and low-pass filtering applied to noisy optical time-domain backscatter signals. Reason: noise spectrum overlapped with signal frequency content, yielding negligible variance reduction
Raman optical time domain reflectometry for aircraft fire-overheat detection and monitoring · Cranfield
Considered and rejected
Considered and rejected: Rejected naive low-pass/band-pass direct inverse filtering due to large noise spikes and high-frequency informational loss, replacing it with wavelet denoising.
NONDESTRUCTIVE TESTING AND MATERIAL CHARACTERIZATION BY TERAHERTZ PULSED IMAGING AND TIME-DOMAIN SPECTROSCOPY · Georgia Tech
Considered and rejected
Considered and rejected: Rejected bandpass filtering because temporal filtering is ineffective at eliminating signals above Nyquist frequency (~0.143 Hz) and aliases physiological noise.
Considered and rejected
Considered and rejected: Rejected using standard moving average filtering (filter function) because it degrades essential geometric information of chaotic trajectories (used wavelet denoising wden instead).
Optimization Algorithm to Determine Parameters and Track Nonlinear Dynamic Systems in Pharmacology · DSpace at SUNY Buffalo
Considered and rejected
Considered and rejected: Rejected low-pass filtering on MEMS-IMU high-frequency white noise because filtering erases essential dynamic information leading to EKF instability; handled via measurement noise tuning instead.
Considered and rejected
Considered and rejected: Rejected conventional Butterworth filtering for limb EMG line noise removal in favor of a time-domain sine subtraction method to avoid distorting mEP waveform morphology.
Optimizing Brain Stimulation for Parkinson's Disease, Memory Enhancement, and Optogenetic Control · Georgia Tech
Tried and failed
broadband filtering for thermal noise reduction applied to time-domain transient electrical signals. Reason: filtering distorted and attenuated the signal due to spectral overlap between noise and fast pulse transients
Time, momentum, spin, and energy resolved tunneling spectrum of a two-dimensional electron system · MIT
Tried and failed
low-pass filtering of sensor signals applied to inertial measurement unit acceleration data. Outcome: did not generalise. Reason: broadband motion frequencies overlapped with noise during dynamic movement phases
Multi-sensor systems and models for high accuracy indoor positioning · Imperial
Tried and failed
moving average filtering applied to power spectral density estimation. Outcome: worse than baseline. Reason: it over-smoothed spectral peaks compared to ensemble dataset averaging
Wind Tunnel Testing to Evaluate Noise Emissions from a Small Wind Turbine · Carleton University Institutional Repository
Tried and failed
brick-wall frequency filtering for displacement estimation applied to tissue motion ultrasound time series. Outcome: worse than baseline. Reason: removes low-frequency physiological movement data, degrading measurement repeatability
Tried and failed
frequency domain filtering to suppress low frequencies applied to image sensor noise artifact extraction. Outcome: worse than baseline. Reason: filtering removed critical low-frequency camera artifact signals, degrading performance to random guessing
NoiseLearner: An Unsupervised, Content-agnostic Approach to Detect Deepfake Images · Virginia Tech
Considered and rejected
Considered and rejected: Rejected aggressive signal pre-filtering prior to the inverse solution because it distorts reconstructed electrogram features compared to ground truth.
Feasibility of improving risk stratification in the inherited cardiac conditions · Imperial
Considered and rejected
Considered and rejected: Rejected relying purely on digital high-pass filtered data (0.005 Hz cutoff) for characterizing spreading depolarizations because filtering distorts event duration and morphology on long-duration events, requiring reversion to unfiltered DC recordings
Quantifying Biomarkers for Brain Disease State Monitoring and Intervention · DukeSpace
Considered and rejected
Considered and rejected: Rejected conventional analog low-pass filtering because it removes high-frequency kinetic and mechanistic components of single-entity events; adopted digital band-stop filtering a posteriori instead.
Advances in Single Entity Electrochemistry for Semiconducting Nanocrystal Studies · unevada
Considered and rejected
Considered and rejected: Aggressive Butterworth filtering (critical freq 0.05) rejected for general preprocessing because it distorted underlying EGM morphology and amplitudes
Predicting electrophysiological function of ex-vivo hearts using machine learning · Imperial
Considered and rejected
Considered and rejected: Rejected temporal downsampling/filtering because it removes high-frequency components and distorts underlying causal structure
Considered and rejected
Considered and rejected: Rejected standard signal-conditioning low-pass filtering on pressure transducer outputs in order to retain high-frequency noise and spectral dynamics for plant diagnostics.
Flow measurement and monitoring using orifice plates · Cranfield
Tried and failed
strict quality filtering of training data applied to spectroscopic deep learning quantification. Outcome: did not generalise. Reason: over-filtering excluded natural variance and outliers, reducing model robustness and generalizability on unseen data
Spectroscopic MRI in the Study and Clinical Translation of Pediatric High-Grade Glioma and Neurologic Disorders · Georgia Tech
Considered and rejected
Considered and rejected: Rejected low-pass moving average and moving median filters for time series smoothing because they attenuated cyclic features like peak height and width, selecting the Savitzky-Golay filter instead.
Considered and rejected
Considered and rejected: Rejected Savitzky-Golay filtering and Bruker baseline correction because they over-smoothed or failed to remove IR artifacts, selecting Whittaker smoothing instead.
Considered and rejected
Considered and rejected: Rejected applying Savitzky-Golay noise filtering because it smoothed out real anomalies, added computational cost, and increased latency.
ANOMALY DETECTION IN A MICROGRID USING MACHINE LEARNING METHODS TO ENHANCE POWER SECURITY · Calhoun
Complex and adaptive filtering techniques fail to outperform simpler baselines or raw signals
Sophisticated algorithms such as neural filters, Kalman estimators, and specialized transform models frequently performed worse than basic moving averages, linear regressions, or raw unfiltered data. In multiple evaluations, processing with complex filters introduced unnecessary error and reduced accuracy compared to standard baseline methods.
Tried and failed
single-trial LSTM prediction applied to somatosensory evoked potential estimation. Outcome: worse than baseline. Reason: high electroencephalogram noise prevented statistically significant improvement over stimulus-averaged baselines
Neurological Disease Diagnosis and Treatment via Precise Robotic Intervention · Georgia Tech
Tried and failed
spectral subtraction and Wiener filtering for denoising applied to transient experimental flow measurements. Outcome: worse than baseline. Reason: Provided no practical improvement over simple ensemble averaging across repetitions.
Vortex formation downstream of an active vane vortex generator · Cranfield
Tried and failed
unscented Kalman filter for motion state estimation applied to pedestrian multi-object tracking. Outcome: worse than baseline. Reason: nonlinear filtering degraded tracking accuracy compared to standard linear Kalman filter models
Improved 2D Camera-Based Multi-Object Tracking for Autonomous Vehicles · Virginia Tech
Tried and failed
standard SVM kernels for cross-correlation applied to weak signal detection in noise. Outcome: worse than baseline. Reason: underperformed compared to linear matched filtering and standard cross-correlation baselines
Enhanced Weak Signal Detection Using SVM Based Correlation Algorithm · Virginia Tech
Lost to a baseline
Threshold control, 2-D quadratic filtering, and CWM filtering failed to outperform standard median filtering on the phantom ultrasound data.
Median Filtering and Wavelet Filtering: A Study of Noise Eeduction in Prostate Ultrasound Images · TXST Digital Repository
Lost to a baseline
20 kHz lowpass filter baseline achieved 25.21 μm standard deviation compared to 25.47 μm for the proposed method on supported calibration points.
HIGH-SPEED ROTOR TIP CLEARANCE MEASUREMENTS IN A TRANSONIC COMPRESSOR · Calhoun
Lost to a baseline
In Scenario 2 linear motion preliminary tests, a simple sliding average filter (window length w=100) outperformed EKF in smoothing ALG. B outputs.
Lost to a baseline
Kalman filter prediction method (185 m/year average RMSE) was beaten by simple linear regression rate extrapolation (128 m/year average RMSE).
Lost to a baseline
Kalman Filter and Moving Average smoothing produced higher TTC MAE (0.81 s and 0.86 s) compared to raw UWB measurements (0.62 s excluding 20 m).
Utilization of Wireless Sensors for Pedestrian Safety Studies · Carleton University Institutional Repository
Considered and rejected
Considered and rejected: Rejected UKF and Sigma-Point Kalman filters in favor of EKF because they did not provide significant accuracy improvements relative to their added computational overhead
Cooperative Multi-Robot Systems for Aquatic Environmental Sensing · EPFL
Lost to a baseline
SNN filter achieved lower F1 score (0.771 on PD1, 0.665 on PD2) than moving average (0.84) and Gaussian (1.0) filtering in LPBF anomaly detection.
A Neuromorphic Approach towards Detection and Control of Features of Interest in Dynamical Systems · Research Repository UCD
Lost to a baseline
Piecewise filtered Fourier reconstruction (mean SSIM 0.83302) lost to unfiltered Fourier baseline (mean SSIM 0.90462) and filtered Fourier (mean SSIM 0.90831) on the 2D synthetic dataset.
On the Epistemology of Gibbs Ringing Reduction Algorithm Performance · Research Repository UCD
Lost to a baseline
In GalOx refolding derivative estimation, raw finite-difference approximation achieved a lower mean error (0.78%) than the chosen Savitzky-Golay filter (1.03%).
Novel soft-sensor applications and mechanistic models for biomanufacturing with Escherichia coli · DSpace-CRIS at TU Wien
Lost to a baseline
STRAY algorithm with Savitzky-Golay/Gaussian filtering lost to unfiltered STRAY on clean Dataset 2 photodiode data (unfiltered BD F1 0.92 vs smoothed F1 0.63-0.79).
In-situ process monitoring of laser powder bed fusion for laser parameter optimisation of Ti-6Al-4V alloy parts with overhang features · Research Repository UCD
Lost to a baseline
Raw FFT coefficients (83% crack / 85% deposit) outperformed Wiener filter deconvolution features (70% crack / 69% deposit) on backpropagation neural networks.
Ultrasonic NDE signal classification on steam generator tubes · Iowa State
Lost to a baseline
Gegenbauer reconstruction (SSIM 0.85003) and Piecewise filtered Fourier (SSIM 0.84592) lost to unfiltered Fourier baseline (SSIM 0.91128) and filtered Fourier (SSIM 0.92145) on 2D MRI datasets with estimated edges.
On the Epistemology of Gibbs Ringing Reduction Algorithm Performance · Research Repository UCD
Considered and rejected
Considered and rejected: Rejected Savitsky-Golay noise-reduction filtering, opting to feed raw multi-channel temporal sensor data directly into the 1D-CNN since 1D-CNNs natively learn temporal features from raw signals.
BAYESIAN CONVOLUTIONAL NEURAL NETWORKS FOR ANOMALY DETECTION IN POWER SYSTEMS · Calhoun
Lost to a baseline
Adaptive RLS filter achieves 'lower overall accuracy than the batch method' at 40 cycles due to recency bias / forgetting factor overfitting.
Modeling and Gait Control for Principally Kinematic Locomotion Systems · JScholarship
Lost to a baseline
Synchronous AEF underperforms a lone passive LC filter at higher harmonics (>1-2 MHz) despite outperforming it at fundamental and low harmonics.
Switch-mode active EMI filtering · UT Austin
Denoising algorithms fail when noise violates stationarity, whiteness, or independence assumptions
Filters designed under white Gaussian or stationary noise assumptions broke down when encountering colored, correlated, speckle, or non-stationary noise. Under these mismatched conditions, matched filters and adaptive estimators lost optimality guarantees, canceled signal energy, and experienced severe performance degradation.
Tried and failed
fixed wavelet scattering transforms applied to motor-imagery EEG signal classification. Outcome: worse than baseline. Reason: pre-determined fixed filter banks cannot adapt to task-specific spectral patterns as well as learnable wavelets
From thought to action: enhancing motor-imagery brain computer interfaces through deep learning · Imperial
Tried and failed
Kalman filtering for signal denoising applied to noisy sensor time-series data. Reason: Enhanced target spikes but failed to attenuate speckle noise.
Application of morphological filtering to defect detection in eddy current wheel inspection signals · Iowa State
Tried and failed
optimal detector tuned to assumed noise correlation applied to constant signal detection. Outcome: worse than baseline. Reason: Minimal performance gain when matched, but severe degradation under parameter mismatch compared to a simple filter
Signal detection in fractional Gaussian noise · Iowa State
Lost to a baseline
In colored (Brownian) noise with chirp signals, polynomial KCC was beaten by standard cross-correlation and matched filter baselines
Enhanced Weak Signal Detection Using SVM Based Correlation Algorithm · Virginia Tech
Lost to a baseline
The Variable Frequency Model (VFM) algorithm filters suppress high-frequency components and noise less effectively than Constant Frequency Model (CFM) filters, leading to higher amplitude and frequency errors on noisy experimental data.
A LEAST ERROR SQUARES TECHNIQUE FOR ESTIMATING THE MAGNITUDE AND FREQUENCY OF A VOLTAGE SIGNAL · HARVEST
Lost to a baseline
Colored noise Kalman filter with augmented random bias state loses optimality guarantee, yielding higher K-L divergence metric Eopt than an ideal Kalman filter with known disturbance input
Aerospace Applications of Noise Covariance Identification with Autocovariance Least Squares · ResearchWorks
Tried and failed
matched filtering on untransformed fluctuating intensity data applied to signals with multiplicative noise. Reason: signal-dependent multiplicative noise is canceled alongside signal information during linear filtering
Tried and failed
matched filter demodulation applied to signals in fractional Gaussian noise. Outcome: worse than baseline. Reason: performance degrades severely as noise correlation increases compared to optimal receivers
Signal detection in fractional Gaussian noise · Iowa State
Tried and failed
Wiener filter denoising applied to non-stationary vibration and flux signals. Outcome: did not generalise. Reason: assumes stationarity which failed under fluctuating supply and environmental harmonic interference
Stray Flux Monitoring and Multi-Sensor Fusion Condition Monitoring for Squirrel Cage Induction Machines · Georgia Tech
Tried and failed
adaptive sliding window regression filtering applied to digitized radiographic images. Outcome: did not generalise. Reason: white Gaussian noise assumption failed for most real-world radiographs
Adaptive regression filter technique for edge detection and enhancement in digitized radiographs · Iowa State
Tried and failed
graph convolutional network on patch-induced graphs applied to spectrally diverse graph patches. Outcome: worse than baseline. Reason: fixed low-pass filtering assumptions conflicted with spectrally diverse graph patches
Towards Open World Graph Learning and Applications · Virginia Tech
Tried and failed
online Kalman filtering for multi-rate sensor fusion applied to longitudinal acceleration estimation. Outcome: unstable. Reason: non-systematic measurement noise from low-frequency observation sources degraded state tracking
Crowd-sourced Road Geometry and Accurate Vehicle State Estimation Using Mobile Devices · DSpace at SUNY Buffalo
Tried and failed
least-mean-square adaptive filtering applied to time delay estimation at low SNR. Reason: receiver noise signals were correlated across channels under low signal-to-noise ratio conditions
Phase Transform Time Delay Estimation to Counteract Spectral Haystacking Effects in Jet Exhaust Flow Measurements · Virginia Tech
Tried and failed
Iterative adaptive minimum-variance filtering with structured covariance applied to partially coherent waveform radar pulse compression. Outcome: did not generalise. Reason: Clutter returns across delays increase clutter subspace rank, exhausting available adaptive degrees of freedom
Non-Orthogonal Waveform Use in MIMO Ground Moving Target Indication Radar · Georgia Tech
Considered and rejected
Considered and rejected: Adaptive filter without an external reference signal, rejected due to insufficient performance in tracking non-stationary distortion.
Noice reduction in control signals of industrial sewing machines using adaptive filtering · Institutional Repository University of Moratuwa
Inverse filtering, deconvolution, and thresholding introduce instability and artifacts
Deconvolution and direct inverse filtering amplified high-frequency noise and caused numerical divergence when system frequency responses contained near-zero values. In addition, heuristic amplitude and outlier thresholding schemes overfit background clutter, tracked outliers, or falsely discarded genuine signal features.
Tried and failed
envelope-domain baseline subtraction applied to signal anomaly detection. Outcome: worse than baseline. Reason: causes nonlinear artifact responses when coherent noise and non-zero baselines are present compared to raw RF subtraction
Estimating the reliability of guided wave SHM systems through modelling · Imperial
Tried and failed
thresholding outliers via moving mean and variance applied to noisy time-series sensor data. Outcome: worse than baseline. Reason: failed to reliably distinguish true signal peaks from measurement outliers compared to median filtering
Tried and failed
iterative deconvolution with higher low-pass cutoffs applied to noisy empirical seismic waveforms. Outcome: did not generalise. Reason: High-frequency noise amplification caused ringing artifacts on real data, unlike synthetic tests.
A comparative study of methods used to compute PdP underside reflection functions · Texas Tech
Tried and failed
source signature deconvolution filtering applied to legacy marine seismic reflection data. Outcome: no signal. Reason: failed to clearly image deep subsurface boundaries compared to stronger deconvolution
Tried and failed
iterative deconvolution for signal separation applied to noisy precursor seismic waveforms. Outcome: overfit. Reason: High iteration counts overfit background noise causing clutter, while too few iterations missed true reflection profiles
Seismic Investigation in Upper Mantle Beneath the Asia Using PP precursor Analyses · Texas Tech
Tried and failed
filtering low-SNR data by parameter estimation uncertainty threshold applied to diffusion MRS metabolite signal estimation. Reason: filtering out high-uncertainty samples biased results toward high-concentration and slowly diffusing species
New insights into rodent brain microstructure and metabolism in hepatic encephalopathy · EPFL
Tried and failed
moving average or polynomial smoothing filters applied to time-series data with outlier noise. Outcome: worse than baseline. Reason: linear smoothing filters track outliers instead of rejecting them, degrading parameter estimation accuracy
Tried and failed
phase derivative thresholding for superresolution wave imaging applied to complex media with interfering scatterers. Outcome: did not generalise. Reason: interfering nearby scatterers distort the phase derivative, causing true scattering locations to be filtered out
Lost to a baseline
TDTV lost to NPT on field seismic denoising because it over-compensated for noise-induced signal loss and heavily distorted the seismic image by retaining excessive non-noise data in the extracted noise difference.
Developing Deep Learning Approaches towards Automatic Seismic Fault Interpretation · Research Repository UCD
Lost to a baseline
Conjugate product phase estimation achieved lower RMSE than MLE (0.5214 m vs. 0.5458 m) at an aggressive 10 dB intensity filter threshold.
Lost to a baseline
On the Beatrice field dataset denoising task, BM3D, MFFCNN, ASPP, MPRNet, and DnCNN lost to NPT by misinterpreting noise as geological signal and retaining it in the output.
Developing Deep Learning Approaches towards Automatic Seismic Fault Interpretation · Research Repository UCD
Considered and rejected
Considered and rejected: Rejected SNR and correlation (COR) filtering from ADV hardware manuals because valid data points are frequently misidentified as subpar.
Hydraulic Characterization of Mounded Gravel Fish Nests: Incipient Motion Criteria and Despiking Acoustic Doppler Velocimeter Data · Virginia Tech
Considered and rejected
Considered and rejected: Rejected standard deconvolution inverse filtering of reflection data because wavelets are band-limited, causing massive deconvolution errors.
Computational Methods for One-Dimensional Scattering in Non-Smooth Media · YorkSpace
Tried and failed
direct inverse filtering of adaptive equalization applied to signal reconstruction and cancellation. Outcome: unstable. Reason: filter frequency response contained zeros or near-zeros causing numerical divergence without regularization
Technologies For Next-Generation Optical Communication Systems · Georgia Tech
Lost to a baseline
For large noise multipliers (sigma in {100, 1000}), MLE Relative Error Metric degrades above 0.5 due to jump filtering while the LS estimator remains constant at 0.37-0.41.
State-space filters suffer from model misspecification and nonlinear divergence
Linear state-space formulations and Extended Kalman filters failed to capture nonlinear dynamics and non-Gaussian dependencies between state variables. Attempts to adapt these models suffered from Taylor series approximation errors, matrix ill-conditioning, and instability under dynamic transition noise.
Lost to a baseline
Classic Kalman filtering (1 GMM component) failed to resolve nonlinear/non-Gaussian parabolic joint dependencies between state variables during data assimilation compared to the GMM-DO filter.
Lost to a baseline
Multivariate linear Gaussian state-space model filtering underperformed Kalman filter (test MSE 0.46 vs 0.2846).
Lost to a baseline
Misspecified linear Kalman filter yielded higher correlations but worse relative standard deviations in non-linear state space setups compared to the partial information filter.
ESSAYS ON MACRO FINANCE · Penn
Lost to a baseline
Manually tuned Extended Kalman filters (EKF 1, EKF 2, EKF 3) were outperformed by the LTV-ALS tuned Extended Kalman filter, which achieved lower K-L divergence metric Eopt and smaller initial transient oscillations
Aerospace Applications of Noise Covariance Identification with Autocovariance Least Squares · ResearchWorks
Considered and rejected
Considered and rejected: Rejected non-linear Extended Kalman Filtering (EKF) in favor of Discrete Linear Kalman Filtering (DKF) in rectangular coordinates to minimize computational complexity
Detecting false data injection attacks against smart grid wide area monitoring systems · DSpace-CRIS at TU Wien
Considered and rejected
Considered and rejected: Rejected the Extended Kalman Filter (EKF) in favor of the Unscented Kalman Filter (UKF) to avoid linearisation approximation errors from Taylor series expansion while keeping computational complexity low.
Identification of nonlinear and time-varying systems under dynamic and seismic excitation · IRIS - POLITO - prod
Tried and failed
Recursive least squares adaptive linear prediction filtering applied to instantaneous frequency tracking of magnetic signals. Outcome: worse than baseline. Reason: Suffered from poor transient response and noise rejection compared to normalized LMS
Tried and failed
offline ridge regression for polynomial filter fitting applied to digital predistortion filter identification. Outcome: unstable. Reason: matrix ill-conditioning prevented functional filter derivation, requiring recursive least squares instead
An implementation of the redirected learning architecturefor digital pre-distortion · Iowa State
Tried and failed
least mean squares adaptive filter applied to time-series feature forecasting. Outcome: did not converge. Reason: constant learning rate and high sensitivity to input feature scaling
Tried and failed
linear adaptive FIR filtering with LMS updates applied to nonlinear dynamic plant identification. Outcome: did not converge. Reason: Linear FIR filters cannot capture nonlinear dynamics regardless of filter length or step size.
A novel Adaptive Filtering approach to Drive File Identification for Service Environment Replication · Virginia Tech
Lost to a baseline
Proposed ML filter performed 'about 10% worse than the filter with Gaussian process noise' in position RMSE for the zero-thrust scenario due to regularization uncertainty inflation.
Bayesian approaches to low-thrust maneuvering spacecraft tracking · UT Austin
Lost to a baseline
Standard SINDy lost to UQ-SINDy (Spike and Slab and Regularized Horseshoe) on damped nonlinear oscillator and synthetic Lotka-Volterra datasets under noise, failing to learn sparse representations and identifying incorrect dynamic models.
Dimensionality Reduction and Sparsity Promotion for Complex Dynamical Systems · ResearchWorks
Filter-induced phase lag and latency degrade tracking and destabilize feedback loops
Temporal smoothing and high-order low-pass filters introduced significant phase delays and latency that disrupted real-time signal detection. In feedback control applications, the excessive phase lag degraded stability margins and destabilized closed-loop control systems.
Tried and failed
cross-correlation impulse response estimation using filtered sensor signals applied to grid frequency dynamics estimation. Outcome: worse than baseline. Reason: High-pass filtering in measurement units distorted signals and introduced phase unsynchronization artifacts
Data-driven modeling and graph learning for power system operations · UT Austin
Considered and rejected
Considered and rejected: Rejected Butterworth and Chebyshev filters for remote pressure signal processing because they introduced unwanted phase distortion
Data-Driven Approaches in Gusty Aerodynamics: Insights from Sparse Surface Pressure Measurements · Queens University Institutional Repository
Considered and rejected
Considered and rejected: Rejected moving average filter for online magnetic field denoising because it averages all frequency components rather than selectively shrinking coefficients below a threshold, introduces window-dependent time delays, and fails to retain GIC-correlated peaks.
Vulnerability Assessment of Power Transformers and Power Systems to Geomagnetic Disturbances · YorkSpace
Tried and failed
moving average window filtering applied to real-time signal detection. Outcome: too slow. Reason: Introduced unacceptable detection phase delays compared to linear-phase Savitzky-Golay filtering.
A Decentralized Privacy-Preserving Data-Driven Methodology for Energy Trading in the Smart Grid · Georgia Tech
Considered and rejected
Considered and rejected: Rejected the use of the Moving Average Window (MAW) filter for real-time invariant feature extraction due to phase delays compared to the Savitzky-Golay filter.
A Decentralized Privacy-Preserving Data-Driven Methodology for Energy Trading in the Smart Grid · Georgia Tech
Tried and failed
autoregressive decoding with time-varying all-pass filter applied to neural speech synthesis. Outcome: unstable. Reason: teacher forcing caused a loading phase discrepancy in time-varying filter parameter prediction
Controllability and Interpretability in Affective Speech Synthesis · EPFL
Tried and failed
digital low-pass filter on sensor feedback applied to PID pressure control loop. Outcome: unstable. Reason: Excessive phase lag introduced by heavy filtering destabilized the feedback control loop with negligible smoothing benefit.
Design and Characterization of a Pressure Controller Test Stand for Soft Pneumatic Actuator Testing · MIT
Tried and failed
increasing low-pass filter order applied to hardware-in-the-loop stability interface. Outcome: unstable. Reason: higher filter order degrades stability margins without systematic data-driven tuning
Data-driven Power Electronic Converter Control Design in Power System Applications · EPFL
Tried and failed
iterative LQR and energy-based nonlinear control applied to stochastic nonlinear dynamical systems. Outcome: unstable. Reason: could not maintain performance or stabilize systems under dynamical Gaussian transition noise
Learning, Optimization, and Control for Real-World Physical Systems · Harvard
Tried and failed
adaptive threshold tracking via peak mean estimation applied to real-time spike detection. Outcome: unstable. Reason: noise peaks reduced estimated peak values and thresholds, creating positive feedback instability without frequent resets
Left open by the authors
Problems the authors named and did not get to.
Left open
Develop a robust observation noise variance estimator for source localization to prevent overestimation from broadband background neural power. Blocker: Lacks a specified algorithmic formulation or mathematical approach for distinguishing cortical observation noise from broadband power
State Space Methods Using Biologically-Relevant Generative Models to Analyze Neural Signals · MIT
Left open
Tune wavelet bases, filter configurations, and KNN hyperparameters in the ECG motion artifact classification and denoising pipeline. Blocker: None
Left open
Evaluate alternative signal embedding methods and optimize density filtering techniques beyond radial histograms for toroidal blind signal separation. Blocker: Vague task specification without defined alternative embeddings, filtering objectives, or benchmark datasets
TOPOLOGY-INSPIRED TECHNIQUES FOR BLIND SIGNAL PROCESSING OF CONSTANT MODULUS SIGNALS · Calhoun
Left open
Develop a principled residual uncertainty tracking method for low-rank Kalman filters to prevent overconfidence without heuristic covariance inflation. Blocker: Lacks a concrete mathematical formulation or specific algorithmic mechanism for tracking truncated residual uncertainty.
Probabilistic Inference for Spatiotemporal Dynamics · Publikationssystem UB Tuebingen
Left open
Validate AutoEKF on synthetic data and benchmark against particle filter methods, or replace linearization with variational autoencoders. Blocker: None
Left open
Generalize filter temporal ablation methods to multichannel EEG deep learning models by implementing per-channel filter perturbations. Blocker: None
Novel Explainability Approaches for Analyzing Functional Neuroinformatics Data with Supervised and Unsupervised Machine Learning · Georgia Tech
Left open
Develop statistical variance models for voltammetric current measurements incorporating baseline drift and capacitive effects. Blocker: Requires experimental voltammetry data with known ground-truth noise profiles or a wet lab to collect calibration datasets
Voltammetric Methods Augmented with Physical Models and Statistical Inference · MIT
Left open
Benchmark RobustICA and adaptive filtering against the proposed template subtraction pipeline for ECG motion artifact isolation. Blocker: None
Motion Artifact Data to Facilitate Bioelectric Signal Quality Analysis Research · Carleton University Institutional Repository
Left open
Test the adaptive denoising algorithm on wearable EMG and EEG recordings under real-world motion noise. Blocker: Requires collecting real-world motion-corrupted wearable EMG wristband and EEG recording data with physical hardware
Decoding peripheral neural correlates of dexterous movements · Imperial
Left open
Evaluate the 2D Gaussian low-pass filtering method on higher-order strain fields and complex experimental DIC gradient datasets. Blocker: Requires experimental DIC measurement data with complex physical strain gradients.
Measurement of linear gradient strain fields at macroscopic and microscopic scale using digital image correlation Messung der linearen Gradientendehnung Felder auf makroskopischer · DSpace-CRIS at TU Wien
Checking a claim in this area?
We can run the same search on any method or claim. If nothing turns up, we will say so, and that proves nothing on its own.