Quaternion recursive adaptive filtering for multidimensional parameter estimation and vehicle localization in non-gaussian environments

IN607662BActive Publication Date: 2026-10-28ATAL BIHARI VAJPAYEE INDIAN INST OF INFORMATION TECH & MANAGEMENT ABV IIITM GWALIOR
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
IN202521069515
Authority / Receiving Office
IN · IN
Patent Type
Patents
Current Assignee / Owner
Filing Date
2025-07-21
Publication Date
2026-10-28
Estimated Expiration
2045-07-21

AI Technical Summary

Technical Problem

Conventional adaptive filtering techniques struggle with quaternion-valued signals and non-Gaussian noise environments, leading to reduced robustness and performance in applications like wireless communications and vehicle localization.

Method used

A quaternion recursive adaptive filtering system employing non-minimum mean square error criteria, such as quaternion recursive maximum correntropy criterion (QRMCC) and quaternion recursive maximum versoria criterion (QRMVC) algorithms, to handle multidimensional parameter estimation tasks effectively in non-Gaussian noise conditions.

Benefits of technology

The system achieves improved convergence speed, reduced steady-state misalignment, and enhanced estimation accuracy in complex, dynamic environments, handling both 3D and 6D pose information with robustness against impulsive and heavy-tailed noise.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

ABSTRACT QUATERNION RECURSIVE ADAPTIVE FILTERING FOR MULTIDIMENSIONAL PARAMETER ESTIMATION AND VEHICLE LOCALIZATION IN NON-GAUSSIAN ENVIRONMENTS The present disclosure provides a quaternion recursive adaptive filtering system, comprising a quaternion input channel (102) configured to receive and process a quaternion-valued input signal; a quaternion adaptive filter (104) coupled to the quaternion input channel (102) and configured to generate a filtered output signal; an error calculation unit (106) configured to compute an error signal based on a difference between the filtered output signal and a desired signal; and a parameter update unit (108) configured to adjust filter coefficients of the quaternion adaptive filter (104). The parameter update unit (108) adjusts filter coefficients using a non-minimum mean square error criterion and implements at least one of a quaternion recursive maximum correntropy criterion (QRMCC) algorithm or a quaternion recursive maximum versoria criterion (QRMVC) algorithm to update the filter coefficients in the presence of non-Gaussian noise. Figure 1
Need to check novelty before this filing date? Find Prior Art

Description

FIELD OF INVENTION

[0001] The present disclosure relates to quaternion-based adaptive filtering system and methodfor multidimensional parameter estimation in signal processing applications. Particularly, thepresent invention discloses a recursive least squares-based quaternion adaptive filtering algorithmsthat employ non-minimum mean square error criteria for robust parameter estimation in non-Gaussian noise environments across wireless communications, vehicle localization, and trajectoryprediction domains.BACKGROUND OF THE INVENTION

[0002] Quaternion-based signal processing techniques have gained increasing attention in variousfields requiring representation and analysis of multidimensional data. Quaternions provide acompact and efficient way to encode information in three or four dimensions, making themparticularly useful for applications such as wireless communications, robotics, computer graphics,and navigation systems.

[0003] In wireless communication systems, quaternion-valued representations are beneficial forprocessing signals with different polarization or spatial dimensions. The use of quaternions in thiscontext may allow for more comprehensive modeling of complex channel characteristics andpotentially improve estimation accuracy in multidimensional scenarios.

[0004] For robotics and navigation applications, quaternions are often employed to representrotations and orientations in three-dimensional space. This representation help ensure smooth andaccurate motion control, as well as facilitate efficient computation of spatial transformations. Inthe realm of vehicle localization and tracking, quaternion-based approaches may offer advantagesin representing and processing pose information including both position and orientationcomponents.

[0005] Traditional adaptive filtering techniques, such as the least mean squares (LMS) algorithmand its variants, have been widely used for parameter estimation and signal processing tasks.However, these conventional methods may face challenges when dealing with quaternion-valuedsignals or when operating in environments characterized by non-Gaussian noise distributions.

[0006] Non-Gaussian noise environments are commonly encountered in practical scenarios,including urban wireless channels subject to impulsive interference or vehicle localization systemsaffected by sensor anomalies and environmental factors. In such cases, filtering algorithms basedon minimum mean square error (MMSE) criteria may not perform optimally, as they primarilyconsider second-order statistics of the error signal.

[0007] In one patent application US 8515712 B1 discloses adaptive filtering techniques for signalprocessing applications, however, the invention is limited to conventional mean square errorcriteria and does not address quaternion-based signal processing or non-Gaussian noiseenvironments. The invention also do not provide the robust performance characteristics needed formultidimensional parameter estimation in challenging noise conditions.

[0008] In another patent application US Patent 8681041 B2 describes adaptive filtering systemsfor various applications. The invention is constrained to conventional adaptation algorithms anddoes not incorporate the quaternion recursive maximum correntropy criterion (QRMCC) orquaternion recursive maximum versoria criterion (QRMVC) algorithms. This limitation of theinvention result in reduced robustness when operating in non-Gaussian noise environments.

[0009] In view of the challenges associated with the above state-of-art, there is a need for anadvanced filtering techniques to effectively process quaternion-valued signals and adapt to diversenoise environments, thereby enhancing the performance across a wide range of applications, fromimproving wireless communication systems to enabling more accurate and reliable vehiclelocalization and tracking capabilities.OBJECTIBVES OF THE INVENTION

[0010] The primary objective of the invention is to provide a quaternion recursive adaptivefiltering system that effectively process multidimensional parameter estimation problems in non-Gaussian noise environments using robust adaptation criteria beyond conventional minimum meansquare error approaches.

[0011] Another objective of the invention is to develop quaternion-based adaptive filteringalgorithms that utilize non-minimum mean square error criteria, such as maximum correntropycriterion and maximum versoria criterion, to achieve improved robustness against impulsive andheavy-tailed noise distributions commonly encountered in practical signal processing scenarios.

[0012] Another objective of the invention is to provide a unified framework for quaternion-valuedsignal processing that handle both three-dimensional and six-dimensional parameter estimationtasks, enabling applications in wireless communications, vehicle localization, and trajectoryprediction with enhanced computational efficiency.

[0013] Another objective of the invention is to implement quaternion recursive algorithms thatexhibit faster convergence characteristics compared to conventional stochastic gradient descentbasedmethods, potentially reducing adaptation time and improving real-time processingcapabilities for dynamic signal environments.

[0014] Another objective of the invention is to enable integrated processing of multidimensionaldata where correlations between signal dimensions may be effectively captured and utilizedthrough direct quaternion domain operations, potentially improving estimation accuracy forcomplex multidimensional signals.

[0015] Yet another objective of the invention is to provide adaptive filtering techniques that handleabrupt changes in signal characteristics, such as sudden vehicle maneuvers or channel variations,through the use of nonlinear error terms in the adaptation algorithms.

[0016] Yet another objective of the invention is to offer flexibility in algorithm selection betweenquaternion recursive maximum correntropy criterion and quaternion recursive maximum Versoriacriterion approaches, allowing users to optimize performance for specific noise environments andapplication requirements.

[0017] Yet another objective of the invention is to achieve reduced steady-state misalignment inparameter estimation tasks, potentially resulting in more accurate long-term tracking andestimation performance compared to conventional quaternion adaptive filtering approaches.

[0018] Other objectives and advantages of the present invention will become apparent from thefollowing description taken in connection with the accompanying drawings, wherein, by way ofillustration and example, the aspects of the present invention are disclosed.SUMMARY OF THE INVENTION

[0019] The present invention discloses a quaternion recursive adaptive filtering system andmethod utilizing non-minimum mean square error criteria for robust parameter estimation in non-Gaussian noise environments. The system is designed for multidimensional signal processingapplications, including channel estimation in wireless communications and real-time vehiclelocalization and trajectory prediction.

[0020] The quaternion recursive adaptive filtering system of the present invention offers severaladvantages, including improved convergence speed and reduced steady-state misalignmentcompared to existing algorithms. It implements novel quaternion recursive maximum correntropycriterion (QRMCC) and quaternion recursive maximum versoria criterion (QRMVC) algorithms,which are particularly effective in handling impulsive and non-Gaussian distortions. The system'sability to process quaternion-valued signals makes the invention well-suited for 3D and 4Dparameter estimation problems, enhancing the applicability in various fields such as robotics,wireless communications, and autonomous vehicle navigation.BRIEF DESCRIPTION OF FIGURES

[0021] The present invention will be better understood after reading the following detaileddescription of the presently preferred aspects thereof with reference to the appended drawings, inwhich the features, other aspects and advantages of certain exemplary embodiments of theinvention will be more apparent from the accompanying drawing in which:

[0022] Figure 1 illustrates a block diagram of a quaternion-based adaptive filtering system;

[0023] Figure 2 illustrates a block diagram of a vehicle localization and tracking system; and

[0024] Figure 3 illustrates graphs showing mean square deviation performance for different noisedistributions for a quaternion-valued channel estimation problem.DETAILED DESCRIPTION OF THE INVENTION

[0025] The following detailed description and embodiments set forth herein below are merelyexemplary out of the wide variety and arrangement of instructions which can be employed withthe present invention. The present invention may be embodied in other specific forms withoutdeparting from the spirit or essential characteristics thereof. All the features disclosed in thisspecification may be replaced by similar other or alternative features performing similar or sameor equivalent purposes. Thus, unless expressly stated otherwise, they all are within the scope ofthe present invention.

[0026] Accordingly, those of ordinary skill in the art will recognize that various changes andmodifications of the embodiments described herein can be made without departing from the scopeof the invention. In addition, descriptions of well-known functions and constructions are omittedfor clarity and conciseness.

[0027] The terms and words used in the following description and claims are not limited to thebibliographical meanings but are merely used to enable a clear and consistent understanding of theinvention. Accordingly, it should be apparent to those skilled in the art that the followingdescription of exemplary embodiments of the present invention are provided for illustrationpurpose only and not for the purpose of limiting the invention.

[0028] It is to be understood that the singular forms "a," "an," and "the" include plural referentsunless the context clearly dictates otherwise.

[0029] It should be emphasized that the term "comprises / comprising" when used in thisspecification is taken to specify the presence of stated features, integers, steps, or components butdoes not preclude the presence or addition of one or more other features, integers, steps,components, or groups thereof.

[0030] Accordingly, the present invention relates to quaternion recursive adaptive filtering systemand method for multidimensional parameter estimation in signal processing applications.Particularly, the present invention discloses a recursive least squares-based quaternion adaptivefiltering algorithms employing non-minimum mean square error criteria for robust parameterestimation in non-Gaussian noise environments across wireless communications, vehiclelocalization, and trajectory prediction domains. By leveraging quaternion mathematics and robustadaptation algorithms, the quaternion recursive adaptive filtering system provide improvedperformance for multidimensional signal processing tasks in non-Gaussian noise environments.The quaternion recursive adaptive filtering system enable more accurate estimation and trackingof parameters represented in quaternion form across various domains requiring analysis of higherdimensionaldata.

[0031] The quaternion recursive adaptive filtering system (100) comprises several componentsworking together to process multidimensional signals and adapt to non-Gaussian noiseenvironments. The system (100) comprises a quaternion-valued input channel (102), a quaternionadaptive filter (104), an error calculation unit (106) and a parameter update unit (108). Byintegrating all the components, the quaternion recursive adaptive filtering system (100) enablerobust multidimensional signal processing and parameter / channel estimation in challenging noiseenvironments. The use of quaternion mathematics and advanced adaptation techniques allow forimproved handling of complex, higher-dimensional data across various applications.

[0032] The quaternion input channel (102) is configured to receive and process quaternion-valuedinput signals. The quaternion input channel (102) handle multidimensional data represented inquaternion form. As illustrated in figure 1, the quaternion input channel (102) is connected to thequaternion adaptive filter (104). The quaternion input channel (102) prepare the quaternion-valuedinput signal for further processing by the adaptive filter.

[0033] In an exemplary embodiment, the quaternion input channel (102) receive input signalsrepresenting multidimensional channel parameters in a wireless communication system. The inputsignal may encode information about channel characteristics includes such as but not limited toamplitude, phase, polarization, and spatial properties in quaternion form.

[0034] The quaternion input channel (102) also process input signals related to vehicle localizationand tracking. The vehicle pose information may be encoded into a pure quaternion representationbefore being received by the quaternion input channel (102). The quaternion encoding maycompactly represent the vehicle's position and orientation in three-dimensional space.

[0035] In another exemplary embodiement, a vehicle's pose comprising x, y, z coordinates androll, pitch, yaw angles may be encoded as a quaternion:Equation;where qw represents the real part and qx , qy , qz represent the imaginary parts of the quaternionand configured as N x 1 vectors. For N = 2, the extended vector parts of the quaternion encapsulatesix degrees of freedom (DoF) associated with the vehicle's pose, enabling a unified mathematicalframework for simultaneous representation of spatial position and orientation.

[0036] The quaternion input channel (102) perform initial processing on the received quaternionsignals. The processing comprises operations includes such as but not limited to normalization,scaling, or coordinate transformations to prepare the input for adaptive filtering. The quaternioninput channel (102) also apply pre-filtering or noise reduction techniques to improve signal qualitybefore passing the processed quaternion signal to subsequent stages of the system (100).

[0037] The quaternion adaptive filter (104) is coupled to the quaternion input channel (102) andconfigured to generate a filtered output signal based on the processed quaternion-valued input. Thequaternion adaptive (104) filter employ various filtering techniques to process themultidimensional data represented in quaternion form. As illustrated in figure 1, the quaternionadaptive filter (104) receive the processed quaternion input signal from the quaternion inputchannel (102). The quaternion adaptive filter (104) further apply adaptive filtering algorithms tothe input signal to produce a filtered output signal.

[0038] In an exemplary embodiment, the quaternion adaptive filter (104) utilize quaternion-valuedfinite impulse response (FIR) or infinite impulse response (IIR) structures. The structures enablethe filter to process the real and imaginary components of the quaternion input simultaneously.The quaternion adaptive filter (104) employ techniques such as quaternion least mean squares(QLMS) or quaternion recursive least squares (QRLS) algorithms to adapt the coefficients. Inanother exemplary embodiment, the algorithms are also extended to handle non-Gaussian noiseenvironments with the help of the use of robust cost functions.

[0039] In another exemplary embodiment, applications involving vehicle localization andtracking, the quaternion adaptive filter (104) are configured to process quaternion-encoded poseinformation. The quaternion adaptive filter (104) are extended to handle 6 degrees of freedom,representing both position (X, Y, Z coordinates) and orientation (roll, pitch, yaw angles) inquaternion form. The quaternion representation for a 6 degree of freedom pose may be expressedas:q = qw + qxi + qyj + qzkwhere qw represents the scalar part and qx , qy , qz represent the vector parts encoding both positionand orientation information.

[0040] In processing the 6 degree of freedom quaternion input, the quaternion adaptive filter (104)apply techniques such as quaternion recursive approaches using non-MMSE criteria. Thesemethods may enable the filter to estimate and track the full 6D pose of a vehicle or object overtime.

[0041] The quaternion adaptive filter (104) also incorporate motion models specific to theapplication domain. In vehicle tracking scenarios, the quaternion adaptive filter (104) may usequaternion-based kinematic models to predict future poses based on current state and controlinputs. By leveraging the quaternion representation and specialized filtering techniques, thequaternion adaptive filter (104) efficiently process complex multidimensional signals and generatefiltered outputs suitable for various estimation and tracking tasks.

[0042] The error calculation unit (106) are configured to compute an error signal based on adifference between the filtered output signal and a desired signal. As illustrated in figure 1, theerror calculation unit (106) may receive inputs from both the quaternion adaptive filter (104) anda desired signal source. The error calculation unit (106) perform a quaternion subtraction operationto determine the difference between the filtered output signal generated by the quaternion adaptivefilter (104) and the desired reference signal. The difference represent the instantaneous error in theadaptive filtering process. The error signal computed by the error calculation unit (106) isexpressed in quaternion form as:e(n) = d(n) - y(n)where e (n) is the quaternion error signal, d (n) is the desired quaternion signal, and y(n) is thefiltered output quaternion signal at time step n.

[0043] In an exemplary embodiment, the error calculation unit (106) apply additional processingto the raw error signal. The error calculation unit (106) compute error statistics includes such asbut limited to mean squared error or higher-order error moments in the quaternion domain. Theerror signal generated by the error calculation unit (106) serve as a key input for the adaptiveprocess.

[0044] The error signal is fed to the parameter update unit to guide the adjustment of filtercoefficients. By minimizing a cost function based on the error signal, the adaptive filtering systemoptimize the performance for the given input and desired output characteristics.

[0045] The error calculation unit (106) process quaternion-valued errors representingdiscrepancies in multiple dimensions simultaneously in applications involving multidimensionalchannel estimation or vehicle localization. The quaternion error formulation enable efficientcomputation and utilization of multidimensional error information within the adaptive filteringframework.

[0046] The parameter update unit (108) is configured to adjust filter coefficients of the quaternionadaptive filter (104). As illustrated in figure 1, the parameter update unit (108) receive inputs fromthe error calculation unit (106) and provide updated filter coefficients to the quaternion adaptivefilter (104).

[0047] The parameter update unit (108) implement non-minimum mean square error criteria toadapt the filter coefficients in the presence of non-Gaussian noise. The parameter update unit (108)utilize specialized algorithms such as the quaternion recursive maximum correntropy criterion(QRMCC) algorithm or the quaternion recursive maximum Versoria criterion (QRMVC)algorithm. The QRMCC algorithm implemented by the parameter update unit (108) use a costfunction of the form:Equation;where e (k) represents the prediction error, λ is a forgetting factor, and σ is a kernel width parameterserving as the shape parameter for the QRMCC algorithm.

[0048] Similarly, the QRMVC algorithm implemented by the parameter update unit (108) employa cost function of the form:Equation;where τ serves as the shape parameter for the QRMVC algorithm.

[0049] In an exemplary embodiment, the parameter update unit (108) initialize algorithmparameters including a number of training observations, a forgetting factor, and the appropriateshape parameter (σ or τ) depending on the chosen algorithm. For each iteration, the parameterupdate unit (108) calculate a quaternion filter output and a prediction error. The parameter updateunit (108) then update a gain vector and a covariance matrix based on the above calculations. Usingthe updated gain vector and covariance matrix, the parameter update unit (108) update the filterestimates.

[0050] The parameter update unit (108) employ different nonlinear error calculations for theQRMCC and QRMVC algorithms upon updating the filter estimates. The nonlinear error termshelp the algorithms to adapt to non-Gaussian noise environments.

[0051] In another exemplary embodiment, the parameter update unit (108) utilize HR-calculus,generalized HR (GHR) calculus, and involution method for gradient computation in the QRMCCand QRMVC algorithms. The mathematical techniques enable efficient computation of quaterniongradients, facilitating the adaptation process in the quaternion domain.

[0052] By implementing the advanced adaptation algorithms, the parameter update unit (108)enable the quaternion adaptive filter (104) to effectively process multidimensional signals andadapt to challenging non-Gaussian noise environments. The continuous adjustment of filtercoefficients based on error feedback and robust cost functions allow the system (100) to maintainoptimal performance across various signal processing applications.

[0053] The quaternion recursive adaptive filtering system (100) operate through coordinatedinteractions between the components to perform multidimensional parameter estimation in non-Gaussian environments. Figure 1 illustrates the overall structure and data flow of the system (100).

[0054] In some cases, the quaternion input channel (102) receive a quaternion-valued input signalrepresenting multidimensional data. This input signal is processed and passed to the quaternionadaptive filter (104). The quaternion adaptive filter (104) generate a filtered output signal basedon its current filter coefficients. The error calculation unit (106) compute an error signal bycomparing the filtered output to a desired signal. The above error signal is then provided to theparameter update unit (108). The parameter update unit (108) uses the error information to adjustthe filter coefficients of the quaternion adaptive filter (104), employing non-minimum mean squareerror criteria to optimize performance in non-Gaussian noise conditions.

[0055] The system (100) construct an augmented quaternion regressor vector from current andpast motion inputs for applications involving vehicle localization and tracking. The augmentedvector incorporate quaternion-encoded pose information and involutions to capture the fulldimensionality of the vehicle's motion history.

[0056] The system (100) estimate a quaternion-valued vehicle pose vector from sensor-derivedobservations and the augmented quaternion regressor vector. Figure 2 illustrates a block diagramof a vehicle localization and tracking scenario, showing the processing paths for both leading andfollowing vehicles. The system (100) comprises two main components: a leading vehicle and afollowing vehicle and each vehicle is equipped with the quaternion recursive adaptive filteringsystem (100). For each vehicle, the system (100) processes sensor inputs to estimate the vehicle'spose in quaternion form.

[0057] The leading vehicle's subsystem comprises sensors to provide raw measurements ofposition and orientation. The measurements are converted into a quaternion representation by thequaternion input channel (102). The quaternion adaptive filter (104) processes the input to estimatethe vehicle's pose. The error calculation unit (106) computes the difference between the estimatedpose and the actual pose derived from high-precision sensors or external references. The parameterupdate unit (108) uses the computed error to adjust the filter coefficients using either the quaternionrecursive maximum correntropy criterion (QRMCC) algorithm or a quaternion recursivemaximum versoria criterion (QRMVC) algorithm, ensuring robust performance in non-Gaussiannoise environments that may result from sensor anomalies or environmental factors.

[0058] The following vehicle's subsystem operates in the manner as the leading vehicle isoperating. The following vehicle also incorporates information received from the leading vehicle.The inter-vehicle communication link allows the following vehicle to adjust its trajectory basedon the leading vehicle's position and movement. The quaternion representation enables efficientencoding and transmission of 6D pose information (position and orientation) between vehicles.

[0059] Both vehicles' systems construct augmented quaternion regressor vectors from current andpast motion inputs. These vectors, along with sensor-derived observations, are used to estimatequaternion-valued vehicle pose vectors. The QRMCC and QRMVC algorithms implemented inthe parameter update unit (108) are particularly effective in handling abrupt vehicle maneuversand other non-Gaussian disturbances in the localization process.

[0060] The system (100) continuously predicts vehicle trajectories based on the estimatedquaternion-valued pose vectors. These predictions can be used for collision avoidance, pathplanning, and to generate early warning signals. By leveraging quaternion mathematics and robustadaptation techniques, the system maintains accurate multidimensional parameter estimates evenas environmental conditions or vehicle behaviors change over time.

[0061] This quaternion-based approach to vehicle localization and tracking offers severaladvantages, including compact representation of 6D pose information, efficient processing ofmultidimensional data, and robust performance in challenging non-Gaussian environments typicalof real-world driving scenarios.

[0062] In an exemplary embodiment, the system (100) may model non-Gaussian noise as amixture of Gaussian and non-Gaussian components. The above approach allow for more accuraterepresentation of complex noise environments encountered in real-world scenarios. The noisemodel is expressed as:Equation;where r(n) represents the noise for a given entity, bn is a binary variable, and rB(n) and rO (n)represent background noise and impulsive noise components respectively.

[0063] The quaternion recursive maximum correntropy criterion (QRMCC) and quaternionrecursive maximum Versoria criterion (QRMVC) algorithms implemented by the parameterupdate unit enable robust adaptation in the presence of these non-Gaussian distortions. The abovealgorithms help the system (100) handle abrupt vehicle maneuvers and other challengingestimation scenarios.

[0064] Figure 3 illustrates the mean square deviation (MSD) performance of various quaternionadaptive filtering algorithms under different noise distributions. The graphs demonstrate how theproposed QRMCC and QRMVC algorithms achieve lower steady-state MSD levels as comparedto conventional approaches across various non-Gaussian noise scenarios.

[0065] In another exemplary embodiment, the quaternion recursive adaptive filtering system (100)is applied for integrated sensing and communication applications. The system (100) predictsvehicle trajectories based on the estimated quaternion-valued pose vectors. The trajectorypredictions is communicated to other vehicles or infrastructure systems for early collisionwarnings or avoidance.

[0066] The parameter update unit (108) continuously refine the filter coefficients based on newobservations and error feedback. The adaptive process allow the system (100) to maintain accuratemultidimensional parameter estimates even as environmental conditions or vehicle behaviorschange over time. By leveraging quaternion mathematics and robust adaptation techniques, thesystem (100) enable efficient processing of complex multidimensional data for various estimationand tracking tasks in challenging non-Gaussian environments.

[0067] The quaternion recursive adaptive filtering system (100) offer several advantages formultidimensional parameter estimation in non-Gaussian noise environments:- The present invention provides more effective filtering in the presence of impulsive orheavy-tailed noise distributions as compared to conventional approaches by utilizing nonminimummean square error criteria.- The present invention reduces computational complexity for certain multidimensionalestimation tasks by allowing compact encoding of 3D or 4D information.- The present invention exhibit faster convergence rates as compared to stochastic gradientdescent-based methods, potentially enabling quicker adaptation to changing signalconditions.- The present invention is applicable to diverse multidimensional estimation problems,including wireless channel estimation and vehicle localization / tracking, potentiallyoffering a unified framework for handling various types of quaternion-valued signals.- The present invention effectively capture and utilize correlations between signaldimensions, potentially improving estimation accuracy for certain types ofmultidimensional data by operating directly in the quaternion domain.- The present invention handles sudden variations in signal characteristics, and is beneficialfor tracking maneuvering objects or estimating time-varying channels with the help of thenon-linear error terms in the adaptation algorithms.- The present invention achieve lower steady-state mean square deviation as compared toconventional approaches, potentially resulting in more accurate parameter estimates overtime.- The present invention offers a choice between quaternion recursive maximum correntropycriterion (QRMCC) and quaternion recursive maximum Versoria criterion (QRMVC)algorithms, thereby allowing users to select the most suitable approach for specific noiseenvironments.- The present invention is capable of processing both 3D and 6D pose information inquaternion form, thereby enabling seamless handling of various multidimensionalestimation scenarios.- The present invention allow for efficient online processing of quaternion-valued signals,and is beneficial for applications requiring real-time parameter updates.

[0068] Features of any of the examples or embodiments outlined above may be combined to createadditional examples or embodiments without losing the intended effect. It should be understoodthat the description of an embodiment or example provided above is by way of example only, andvarious modifications could be made by one skilled in the art. Furthermore, one skilled in the artwill recognize that numerous further modifications and combinations of various aspects arepossible. Accordingly, the described aspects are intended to encompass all such alterations,modifications, and va riations that fall within the scope of the appended claims.

Claims

1. A quaternion recursive adaptive filtering system (100), comprising: - a quaternion-valued channel (102) configured to receive and process a quaternionvalued input signal; - a quaternion adaptive filter (104) coupled to the quaternion input channel (102) and configured to generate a filtered output signal; - an error calculation unit (106) configured to compute an error signal based on a difference between the filtered output signal and a desired signal; and - a parameter update unit (108) configured to adjust filter coefficients of the quaternion adaptive filter (104), wherein, - the parameter update unit (108) adjust filter coefficients using a non-minimum mean square error criterion; - the quaternion-valued channel (102) is estimated via proposed a quaternion recursive maximum correntropy criterion (QRMCC) algorithm and a quaternion recursive maximum versoria criterion (QRMVC) algorithm; and - the parameter update unit (108) is further configured to implement at least one of a quaternion recursive maximum correntropy criterion (QRMCC) algorithm or a quaternion recursive maximum versoria criterion (QRMVC) algorithm to update the filter coefficients in the presence of non-Gaussian noise.

2. The system (100) as claimed in claim 1, wherein the parameter update unit (108) is configured to implement the QRMCC algorithm using the following cost function: J(en) = Σk=0 to n λ(n-k) exp(-|e(k)|2 / (2σ2)) where e(k) is a prediction error, λ is a forgetting factor, and σ is a kernel width parameter.

3. The system (100) as claimed in claim 1, wherein the parameter update unit (108) is configured to implement the QRMVC algorithm using the following cost function: J(en) = Σk=0 to n λ(n-k) (1 / (1 + τ|e(k)|2)2) where e(k) is a prediction error, λ is a forgetting factor, and τ is a scale parameter.

4. The system (100) as claimed in claim 1, wherein the quaternion input channel (102) is configured to process a quaternion-valued input signal representing a 3D or 4D parameter estimation problem.

5. The system (100) as claimed in claim 4, wherein the 3D or 4D parameter estimation problem comprises at least one of: multidimensional channel estimation in wireless communication, or real-time vehicle localization and trajectory prediction.

6. A method for quaternion-based adaptive filtering system (100), comprising the steps of: - receiving a quaternion-valued input signal at the quaternion input channel (102); - processing the quaternion-valued input signal using the quaternion adaptive filter (104) to generate a filtered output signal; - computing an error signal based on a difference between the filtered output signal and a desired signal using the error calculation unit (106); and - updating filter coefficients of the quaternion adaptive filter (104) using the parameter update unit (108), wherein, - updating the filter coefficients comprises using a non-minimum mean square error criterion and implementing at least one of a quaternion recursive maximum correntropy criterion (QRMCC) algorithm or a quaternion recursive maximum Versoria criterion (QRMVC) algorithm to adapt the filter coefficients in the presence of non-Gaussian noise.

7. The method as claimed in claim 6, further comprising: - initializing algorithm parameters including a number of training observations, a forgetting factor, and a shape parameter; - calculating a quaternion filter output and a prediction error for each iteration; updating a gain vector and a covariance matrix; and - updating filter estimates based on the updated gain vector and covariance matrix.

8. The method as claimed in claim 7, wherein the shape parameter is σ for the QRMCC algorithm and τ for the QRMVC algorithm, and wherein updating the filter estimates comprises using different nonlinear error calculations for the QRMCC and QRMVC algorithms.

9. The method as claimed in claim 6, further comprising: - processing the quaternion-valued input signal to estimate multidimensional channel parameters in a wireless communication system; and - adapting the quaternion adaptive filter (104) to compensate for non-Gaussian noise in the wireless channel.

10. The method as claimed in claim 9, further comprising: - encoding vehicle pose information into a pure quaternion representation; - constructing an augmented quaternion regressor vector from current and past motion inputs; and - estimating a quaternion-valued vehicle pose vector from sensor-derived observations and the augmented quaternion regressor vector in the presence of non-Gaussian distortions modeling abrupt vehicle maneuvers.