Foundations of Active Control - Active Noise Reduction Helmets

In: Elmkjær , T H L 2008 , Foundations of Active Control - Active Noise Reduction Helmets .

Open Access

Abstract

Denne Ph.D. afhandling omfatter fundamentale betragtninger omkring topologier, algoritmer, implementeringer, metoder etc., der kan indg˚a i næste generation af aktive kontrol systemer. Specifikt foresl˚as der en variant af feedforward kontrol refereret til som indesluttet feedforward aktiv kontrol forkortet IFFAK. I denne topologi indg˚ar et sæt reference sensorer, der er positioneret p˚a en overflade, der fuldt ud indeslutter de ønskede stille-zoner, hvori et sæt performance sensorer monitorerer den opn˚aede støjreduktion. Denne indesluttet-feedforward aktiv kontrol (IFFAK) topologi er indlejret i et mange-input-mange-output (MIMO) system, der omfatter b˚ade feedforward og feedback kontrol. Det totale system er refereret til som et hybrid MIMO indesluttetfeedforward FBS (HMIMOIFFFBS). Undersøgelsen af et komplekst multi-kanals aktiv støjreduktion (ASR) system med hybrid feedforward og feedback topologier er motiveret ud fra krav om høj aktiv støjdæmpning i ekstreme støjmiljøer, som f.eks. opleves ombord p˚a luftb˚arne militære platforme. Støjoptagelser erhvervet ombord p˚a s˚adanne fartøjer afslører lydtryk, der ofte overstiger 140 dB re. 20 μPa. Endvidere udviser disse støjsignaler store tidslige s˚avel som spatiale variationer. Naturlige begrænsninger i feedback baserede aktiv kontrol (AK) systemer som typisk anvendes i moderne ASR støjværn, hvor modstøjssignalet notorisk er forsinket i forhold til den primære forstyrrelse, sætter en øvre grænse for, hvor stor en aktiv dæmpning, der kan opn˚aes. S˚aledes, hersker der i krævende militære applikationer et krav om nye mere avancerede og effektive ASR løsninger. Den opn˚aelige ASR i et FFS er i stor udstrækning bestemt af kohærensen mellem sættet af reference sensorer og sættet af fejl- eller performance sensorer. S˚aledes omfatter denne afhandling en del kohærensundersøgelser baseret p˚a diffustfeltsm˚alinger i et støjkammer samt m˚alinger, der er foretaget i en CH-47D Chinook helikopter. Fra disse kohærensanalyser kan det konkluderes, at IFFAK systemet anvendt p˚a pilothjælme giver mulighed for ca. 25 dB støjreduktion ved 100 Hz faldende til ca. 10 dB dæmpning ved 900 Hz. Endvidere, er der ikke nogen umiddelbare tegn p˚a en mætning med stigende antal reference sensorer. S˚aledes vil et større antal reference sensor forventeligt kunne øge den øvre ASR frekvensgrænse for systemet, der bestemmes af den rumlige samplingstæthed. I hybridsystemet indg˚ar der b˚ade et kontinuerlig-tids FBS og et diskret-tids FBS. Disse vil bidrage med yderligere støjreduktion primært overfor bredb˚andet støj henholdsvis overfor periodiske signaler. Tidsforsinkelser udgør en anden bestemmende faktor for den opn˚aelige effekt i et FFS design, men specielt i et FBS design, eftersom fysiske systemer altid opererer kausalt. For at vurdere størrelsesordenen af det tidsforspring som hver reference sensor giver i forhold til hver fejlsensor i det foresl˚aede IFFAK system indføres en størrelse, der betegnes som den spatialt-vægtet middeltidsgevinst. Der eksisterer imidlertid et problem, n˚ar man forsøger at modellere et fysisk system med en endelig rummelig udstrækning og hvor der s˚aledes ikke er nogen indlysende input-output definition med et endeligt-element multi-kanals system. Som regel eksisterer der ikke nogen unik overføringsfunktion eftersom systemet ikke bliver punktvist stimuleret, men derimod stimuleret xii over et areal som for eksempel under diffustfelts belysning. En ny akustisk signalbehandlingsmetode, der betegnes som samlet kanal residual spektral analyse (SKRSA) er udviklet. Denne metode benyttes til ekstraktion af fælles signal information fra forskellige observationspunkter i rummet. Ideen er at separere hvert spektrum i et kohærent spektrum og et residual spektrum. Indholdet i det kohærente spektrum kan opn˚aes som en linear kombination af spektrene fra de andre kanaler, hvorimod indholdet af det residuale spektrum er unikt for den p˚agældende kanal. I et specifikt eksempel, belyses et system best˚aende af et sæt reference sensorer monteret p˚a en Gentex HGU-55/P hjælm, der igen er p˚amonteret en hoved og torso simulator med et diffust lydfelt. Under anvendelse af SKRSA metoden estimeres den spatialt-vægtet middeltidsgevinst til at være i størrelsesorden 800-900μs. Afhandlingen omfatter ogs˚a en detaljeret beskrivelse af en ny id´e til en beregningsmæssig effektiv implementering af et multi-kanals system, hvor b˚ade de adaptive filtre, der indg˚ar i den aktive kontrol s˚avel som de adaptive filtre, der indg˚ar til modellering af systemoverføringsfunktionerne kan antage individuelle længder. En ny og mere generel variant af APA algoritmen er udviklet. Denne adaptive filter algoritme inkluderer parametre for b˚ade wægt-styret og kontrol-effekt-styret lækage, adaptiv tap-vægte regulering s˚avel som numerisk regulering og betegnes MC-αγΠ-APA. En simplificering af denne algoritme, fører til MC-αγΠ-NLMS algoritmen, der er en udbygget variant af NLMS algoritmen. Systemets evne til off-line simultant at kunne identificere et complex system best˚aende af ialt 28 enkelt systemgrene bliver demonstreret. Forskellig adaptive filtre samt parametering heraf bliver udforsket. Et nyt og generelt multi-hastigheds systemkoncept for aktiv kontrol er udviklet. Specifikt implementeres og testes et system, hvor der i alt samples med tre forskellige hastigheder. P˚a multi-hastighedsniveau 0 benyttes en meget høj samplingsfrekvens med henblik p˚a at reducere forsinkelser i konverteringstrinene, der indg˚ar i de sekundære grene. Den ikke adaptive kontrol udføres p˚a det lavere multi-hastigheds niveau 1. Herved tilsikres et kompromis imellem forsinkelser til afgivelse af modstøjssignaler og krav til en endelig system b˚andbredde. Sluttelig foreg˚ar den adaptive kontrol ved det lavere multi-hastigheds niveau 2. Herved begrænses den ofte beregningsmæssige tunge adaptive filter opdatering til en s˚a lav samplingsfrekvens som muligt. I et specifikt eksempel demonstreres, at en beregningsmæssig besparelse p˚a ca. 40% kan opn˚as under opretholdelse af samme ASR ved nedsampling fra multi-hastighedsniveau 1 p˚a 24 kHz til multi-hastighedsniveau 2 p˚a 3 kHz. Det er en almindelig ingeniørpraksis at foretage en antagelse om Gaussisk fordelte signaler. Imidlertid, er mange fænomener i dagligdagen bedst modelleret med s˚akaldte alfa-stabile fordelings funktioner. Dette gælder ogs˚a for støjsignaler, der ønskes undertrykt ved hjælp af et aktivt støjdæmpningssystem. Afhandlingen indholder en kort teknisk beskrivelse af de stabile fordelingsfunktioner samt adaptive filter algoritmer for disse type signaler. Store dele af HMIMOIFFFB systemet samt de udviklede metoder og algoritmer er implementeret i et realtids miljø, der inkluderer en signal processor. I første omgang vil disse blive aftestet p˚a en til form˚alet designet aktive kontrol testenhed. ; This Ph.D. thesis includes fundamental considerations about topologies, algorithms, implementations, methods etc., that can enter in the next generation of active control (AC) systems. Specifically, a new variant of feedforward control referred to as confined feedforward active control (CFFAC) is proposed. This topology is constituted from a set of reference sensors that are positioned on a surface that completely confines the desired zones of quite. A set of performance sensors monitors the achieved noise reduction. This CFFAC topology in turn is embedded in a multiple-input and multiple-output (MIMO) system that facilitates both feedforward and feedback control. The general system is then referred to as hybrid MIMO confined-feedforward feedback (HMIMOCFFFB) active noise reduction (ANR) system. The investigation of a multi-channel ANR system with hybrid feedforward and feedback topologies is motivated by requirements of high ANR attenuation in extreme noise environments as typically experienced onboard airborne military platforms. Noise recordings acquired on such platforms reveal very high sound pressure levels often exceeding 140 dB re. 20 μPa. Moreover, these noise signals exhibit large temporal as well as spatial variations. Inherent limitations are related to the use of stand-alone feedback AC implementation commonly applied in modern ANR headset. In such systems the anti-noise signal is notoriously behind the primary disturbance in time. Accordingly, in demanding military applications requirements on more advanced and effective ANR system designs prevail. The achievable ANR performance in a feedforward system (FFS) is to a large extent determined by the degree of coherence between the set of reference sensors and the set of error sensors (or performance sensors). Accordingly, this thesis includes a number of coherence analysis that are based on diffuse sound field measurements in a reverberant chamber and measurements conducted onboard a CH-47D Chinook helicopter. From these coherence analysis it can be concluded that the CFFAC system with 10 reference sensors applied to pilot helmets potentially provides approximately 25 dB noise reduction at 100 Hz decreasing to approximately 10 dB attenuation at 900 Hz. Moreover, there is no apparent sign of saturation of the noise reduction with an increasing number of reference sensors. Accordingly, by using more reference sensors the spatial sampling rate is increased which in turn most likely also will lead to an increased ANR bandwidth. The hybrid system is also constituted from a continuous-time feedback system (FBS) and a discrete-time FBS. The continuous-time FBS is primarily responsible for additional broadband noise reduction, whereas the discrete-time FBS primarily is responsible for the attenuation of periodic signals. Owing to the requirement on causal operation of a physical AC system time delays will also to a large extent determine the achievable performance in FFS design and in particular in FBS design. A quantity referred to as the spatially-weighted-averaged acquisition lead time is introduced to represent the averaged time-advance obtained by each reference sensor relative to each performance sensor involved in the proposed CFFAC system. A problem exist when one attempts to model a physical spatially distributed system with no obvious input and output channel definition by a finite lumped-elements multi-channel system. Usually, no unique transfer function x exist as the system is not point-wise excited, but excited over an area as in the case of diffuse sound field illumination. A new method for acoustical signal processing that is referred to as joint-channel residual spectral analysis (JCRSA) is developed. The JCRSA method is used for the extraction of joint signal information from different observation positions in space. The idea is to separate each spectrum in a coherent spectrum and a residual spectrum. The contents of the coherent spectrum can be obtained from a linear superposition of the other signals, whereas the residual spectrum bears information that is unique to each specific channel. In a specific example a system consisting of 10 reference sensors flush-mounted on a Gentex HGU-55/P helmet that in turn is mounted on a head and torso simulator (HATS), is exposed to diffuse sound field illumination. By applying the JCRSA method the spatially-weighted-averaged acquisition lead times provided by the reference sensors relative to the performance sensors are estimated to be as much as 800-900μs. The thesis also includes a detailed description of a new idea for a computational efficient implementation of a multi-channel system in which the adaptive filters for adaptive control as well as the adaptive filters used for plant modeling are allowing to take different lengths. A new and more general variant of the affine projection algorithm has been developed. This adaptive filter algorithm that is denoted by multiple-channel-αγΠ-affine projection algorithm includes parameters for both weight-driven and control-effort-driven leakage, adaptive tap-weight regularization as well as numerical regularization. A simplification of this algorithm leads to the MC-αγΠ-NLMS algorithm that is an extended variant of the NLMS algorithm. Off-line simultaneous system identification capabilities of a complex system involving a total 4 secondary paths, 20 feedback paths and 4 control-performance paths is demonstrated. Different adaptive filters and parameterizations hereof are examined. A novel and general multi-rate adaptive filter for adaptive AC has been developed. Specifically, a system involving 3 different sampling rates has been implemented and the results hereof are presented. In this multi-rate system conversion take place at highly oversampled rates in order to reduce the delays in the secondary paths. The non-adaptive control is performed at a somewhat lower rate. Hereby, a compromise between delays related to the generation of the anti-noise signal and the computational load involved is ensured. Finally, the adaptive control that might be computational intensive takes place at an even slower sampling rate hereby relaxing the requirements on a high bandwidth. It is demonstrated that computational savings as high as 40% can be achieved in a 192, 24, 3 kHz triple-rate system as compared with a 24 kHz single-rate system without sacrificing the ANR performance. It is common engineering practice to apply an assumption of Gaussian distributed signals. However, many phenomena encountered in daily life fall into a generalization of the normal distribution that is referred to as α-stable distributions. Noise sources encountered in the domain of AC are sometimes best fitted to the family of α-stable distributions. This thesis includes a brief technical introduction to the stable distributions and description of the adaptive filter that can be used for AC. Large parts of the HMIMOCFFFB system including the developed methods and algorithms have been implemented in a real-time environment (RTE) that includes a signal processor. Test on the helmet system will continue and a dedicated reference test unit (RTU) for AC is currently being designed.