US 7,184,556 B1 Page 2 6.252,968 B1 * 6/2001 Narasimhan et al. ....... 381/103 6,295,364 B1* 9/2001 Finn et al. ........... .... 38.1/110 6,317, 117 B1 * 11/2001 Goff ........................... 345/156 Toole, Floyd E. and Olive, Sean E., “The Modification of Timbre by Resonances: Perception and Measurement” J. Audio Eng. Soc., vol. 36, No. 3, Mar, 1988, pp. 122-123. Fasoldt, A., Jul/Aug. 1999 column in Fanfare Magazine, pp. 6,721,428 B1 383-384. |U.S. PATENT DOCUMENTS 2003/0132956 A1 4/2004 Allred et al. 7/2003 Duncan Herschelmann, Russ, “Home Theater Architect–Is It Golden or OTHER PUBLICATIONS Goofy?” Stereophile Guide to Home Theater Nov. 1998, pp. 42-44. Mendonsa, Ruth A., “Laser Vibrometer Heightens Speaker Design” Photonics Spectra, Jan. 1998, p. 18. * cited by examiner U.S. Patent US 7,184,556 B1 Sheet 4 of 15 Feb. 27, 2007 F/G, 5 | || NT ????? TNT |N|| || Nº || | HLINII]XJ EXCURSION ?[WITZ N NS ITI-S… O ||| NUH [ XN =U= ENCY IN CYCLES PER SECOND 100 U.S. Patent Feb. 27, 2007 Sheet 6 of 15 `s s g . à ??? US 7,184,556 B1 N U.S. Patent Feb. 27, 2007 Freq Cut/Boost 334405Hz Sheet 7 Of 15 US 7,184,556 B1 Freq Cut/Boost BAD CD NO DONUT Main Compensation#2 | Main || | Smiley Foce 334405Hz BAD CD NO DONUTT || || T U.S. Patent Feb. 27, 2007 Sheet 8 of 15 US 7,184,556 B1 Resonance Compensation;2|Standing Wave rejection | Moin. Tº TT Smiley Foce | Cutoff Response Main I TI Similey Fice Eutoff Response Resonance Compensation #2 istanding Wave Rejection Compensation o Cut/Boost 525846 EAD CO NO DONUTI I T U.S. Patent Feb. 27, 2007 Sheet 9 of 15 US 7,184,556 B1 First Notch Freq Q Cut/Boost Freq Cut/Boost Freq Q Cut/Boost BAD CD NO DONUT Main || || Smiley Foce ICutoff Response Response Compensation Resonance Compensation #2 jStanding Wave Rejection|Allpass|Double Tuned Notch Allpass BAD CD NO DONUT || T U.S. Patent Feb. 27, 2007 Sheet 10 of 15 R F/G. 9 ? US 7,184,556 B1 R2 ? *p p OUTPUT U.S. Patent Feb. 27, 2007 F/G 11 Sheet 11 of 15 US 7,184,556 B1 U.S. Patent Feb. 27, 2007 Sheet 12 of 15 US 7,184,556 B1 LOW ··?? BOOST HIGH "Q” NOTCH COMBINED RESPONSE INPUT OUTPUT LOW Q F/G, 13 U.S. Patent Feb. 27, 2007 Sheet 13 of 15 US 7,184,556 B1 ------- FREQUENCY ADJUST DOUBLE TUNED LOW "Q" BOOSIS DOUBLE TUNED HIGH "Q” NOTCHES ––– FREQUENCY ADJUST COMBINED | RESPONSES | | | H- FREQUENCY ADJUST INPUT (2) -. - G.) OUTPUT S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S SA S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S S FREQUENCY ADJUST F/G, 14 U.S. Patent 2^ w— 0 | 408| ?GO!09 | G 07 09 t<º?-= ?|| ?:_ ?O, ?º000LL00K0L0S0SS00L00LS0SSL0L0S og U.S. Patent Feb. 27, 2007 10? 10K Sheet 15 of 15 10K US 7,184,556 B1 10K R, @ CW, CCW US 7,184,556 B1 1 2 for a wider frequency range can be designed and imple mented in this manner. Several components are needed to COMPENSATION SYSTEM AND METHOD FOR SOUND REPRODUCTION match the resonant behavior, but all interact with each other RELATED APPLICATION This application claims the benefit of co-pending provi sional application Ser. No. 60/148,412 filed Aug. 11, 1999. BACKGROUND OF THE INVENTION 10 1. Field of the Invention This invention relates generally to a compensation method and system for use in sonic transmission and repro duction systems and more particularly to a compensation method and system that uses parametric values to control or adjust processes having transforms or models with proper ties or responses like the components or elements used in the transmission or reproduction system. 2. Description of Related Art Most audio reproduction systems use electromechanical loudspeakers to acoustically reproduce audio signals. The electrical, mechanical, and acoustical properties of the loud speakers are often less than ideal, causing distortions, response anomalies, and other coloration of the sound. Many techniques are used to compensate for the loudspeaker’s characteristics in order to improve perceived audio quality. Functional or behavioral models of loudspeakers are used in practice and found in literature to develop such compen sations. A good example of models and how the modeling process works is described in “Active Equalization of Loud speakers”, Speaker Builder, February 1997. Models consoli date technical languages and are usually intended to imitate or simulate the acoustic responses of the speaker system from electrical stimulus. Model creation or synthesis fre quently begins by making functional groupings of elements which collectively represent or behave like all or part of the speaker. Coil and magnet parts become motors, which are represented by resistors, inductors, capacitors, back EMF generators and other transformed parts. A combination of factors such as air volume, moving mass, acoustic loading, magnetic-braking, and mechanical losses might be analyzed and simplified to LCR resonator networks or circuits. Most 15 20 25 30 35 40 often, the transformed electromechanical, acoustic, and mechanical representations expressed in the model are fur ther simplified or reduced to fewer elements. The model still responds like the speaker, but the parts making the model no longer have exact behavioral equivalence to the parts mak ing the speaker. Consequently, traditional models are neither intended, nor capable of making parametrically addressed zero-phase compensations when speaker parts are changed. One could characterize and invert the frequency response, as well as other properties, of a well-conceived model and achieve linear-phase correction of the loudspeaker. The technique does work to a fashion, but its dedicated, inflex ible circuitry or specialized process tied to the traditional model limits its use to a one-speaker design. Some high quality crossover networks constructed to divide the signal spectrum amongst multiple drivers may have some conju gate response correction like this. A low-frequency resonant boost is intentionally designed for most speakers. Frequently, traditional models are made to represent quantifiable and predictable acoustic behavior as well as other speaker design factors affecting bass response. Mechanical construction and properties of air determine frequency, resonant losses and the configuration’s effect on acoustic output from the speaker. A good approxi mation to a zero-phase conjugate or same-order correction 45 50 when adjustments are made for a different speaker of similar concept and design. Therefore, the operation is not strictly parametrically controlled, as the adjustments must be re calculated from the model to create the minimum-phase or exact match needed for best fidelity with the new speaker. When more corrections are added the interaction problem becomes formidable. The system must be tuned experimen tally or the model analyzed each time an adjustment is made. Consequently, lumped model processes for response flatten ing are inherently designed for a specific speaker. The process must be redesigned for other speakers. Traditional curve-fitting methods can require hundreds of data points and corresponding adjustments to set up and many components or much processing power to match the acquired frequency response. Analog methods are imprac tical and digital processes require much computation and extensive architectures to do this. Neither can provide phase accurate responses or the hidden corrections described later without having knowledge of the speaker and its operation. Without a model, the effort to combine amplitude, time and phase corrections together from measured responses becomes formidable. Some of the most important behaviors of loudspeakers (with respect to acoustically perceptible effect) cannot be modeled or implemented from traditional methods. Such behaviors include standing wave interference, modal breakup, and coupled resonance as well as nonlinear con sequences from such potentially interacting acoustic and mechanical behaviors. Counterproductive random motion or breakup may occur. Even when the average response remains flat or is the same as other frequencies being reproduced, energy can build up during signal stimulus and be released when the signal changes or ceases. In addition, other spatial factors related to stiffness of moving parts and high frequency de-coupling for motions away from a driving voice coil need to be considered. Any of these can create source movements, delayed energy release, and phase error to binaural hearing. Often, such destructive responses can be invisible or very difficult to interpret from traditional micro phone-and-spectrum-analyzer calibration methods. For example, unwanted responses arising from nodal and standing wave behavior affect the settling time, directional behavior, and radiated output of a speaker. Frequently, these responses cause perceptual changes to intelligence signals yet may not be visible or recognizable from response plots. Mechanical motions having large stored energy can be out of phase at different parts of the transducer. The acoustic output might appear to be flat, but human binaural hearing can localize the behavior to its source and the altered ss 60 65 perception can degrade stereo imaging. Often, mechanical disturbances are audible yet invisible or hard to interpret from response measurements made using a frequency sweep and microphone. Parts of a radiating surface can vibrate with different phase relationships to other parts, so that their additive acoustic output is low compared to motions within the transducer and the energy storage involved. When signals at the node frequency change and suddenly stop the release of stored energy can interact with other signals at different frequencies. The resulting beat sounds between the two frequencies can be audible and very objectionable. Sounds with spectra in the interference frequency range may appear louder and granu lar. Human binaural hearing can localize the disturbance to the driver or surfaces from which directional lobes might US 7,184,556 B1 3 bounce, thereby imparting further damage to the stage illusion from multi-speaker stereo reproduction. For this situation, frequencies creating the mechanical disturbance must be sufficiently attenuated to prevent unmasked repro duction of consequential responses. Experience has shown that a sharp, deep notch needed to do this removes enough energy around the correction frequencies to cause a nasal sound. If this inappropriate correction is modified to achieve flat response, then the mechanical sounds remain along with a potential undesirable balance aberration. Many small loudspeakers are constructed with a trans 5 10 ducer, enclosure, and some resonant means to extend bass response such as a port or passive radiator. Usually, these parts are designed to achieve a practical and economical compromise between efficiency, frequency response accu racy, bass extension, and acceptable distortion. Designers of inexpensive, low-powered systems generally opt for higher efficiency to reduce amplifier requirements along with related costs of power supplies and packaging. The com promise situation exposes many undesirable behavioral 15 nodal distortion corrections could be made from such a 20 aspects. Most traditional speaker correction methods apply some variation of amplitude equalization to flatten and extend response from speakers. Adjustments are sometimes done by ear. To be quantitative, one must acquire relevant data. The most common techniques to do this use spectral analysis from noise stimulus. Then, response plots or displays indi cate how an equalizer is to be adjusted. More sophisticated techniques based on delayed acceptance or sampled win dows can measure first-arrival responses from the speaker and remove higher-frequency room disturbance to create anechoic-like data. The intent is to capture information 25 30 relevant either to a listener in a room or to standard mea surement practice where a test microphone is usually speci fied and placed one meter from the speaker. Such technique creates a response that may sound balanced to the single point test microphone. One or more known systems go slightly beyond this by adjusting path lengths, or time delays to align multiple speakers to a listening position. Other techniques provide transient response waveforms, waterfall or successions of spectral plots after an event. Group delay and time-related information is acquired. Such data needs interpretation and has limited use for frequency response leveling practice. Some behavioral responses can be recognized but much more information must be known about the speaker. Measurement devices such as accelerom eters, differential acoustic probes, as well as microphones, are needed for this. Instrumentation may be placed near a suspected behavior site and moved to explore how a response changes with position. Weighted notches can be tuned or slowly swept through suspected frequencies while subjectively observing noise production. More information is needed about dimensions of parts, listening positions, as well as floor, shelf, possibly a computer monitor, or other interceding objects that may be part of the listening envi ronment. Other technical specifications or expressions are needed to complete the conjugate model capable of time phase-accurate correction. A human operator can assume an alignment role by adjusting a graphic equalizer, manually tuning a parametric filter, or changing settings to a crossover device. Commer cial analog components perform these functions, but they have limitations. Graphic equalizers have up to 31 bands or resonators, parametric devices include several adjustable filters and a few have variable crossover and shelf functions. Many more filters are needed. Combinations of graphic and parametric equalizers are incapable of providing a large 4 enough number of points, nor the exact phase and time response to effectively compensate complex behavior from a loudspeaker. Either the corrections do not match specific frequencies, thereby creating phase error, or the number of filters is inadequate to deal with settling time and standing wave issues. Group delay distortion, time-phase error, incomplete correction and other shortcomings are likely to outweigh other improvements. DSP filters can create many more filter sections than is practical from analog circuits. Graphic equalizers made up with parametrically controlled sections have been used with specialized control-generating software to create room response leveling. Such processes are difficult to set up because the room interferes with the identification of impor tant behavioral indicators. Without their input, conjugate response corrections are not possible. Standing-wave and 35 40 45 50 system. However, the awkward compiling and processing needed to parametrically move the compensated notches would be difficult. Most likely, a single point response pickup and FFT has been used for data input to the system. Such methods cannot respond to or provide the time-phase information needed to create a true conjugate response to the speaker. Analysis systems, such as MLSSA, can remove room interference from measurements, and can produce frequency, transient, and settling response data from a loudspeaker system. However, the large amount of data from these measurements must be interpreted. The multiple-band graphic equalizer is not a good choice to install the correc tion. DSP systems can economically create many parametric filters and time-related processes that are impractical with analog circuitry. Traditional large-scale DSP systems have little means to identify and cull out speaker behavior from other measurement anomalies. Their frequency-domain responses are likely to add phase errors and to overlook delayed settling energy. The sound might improve for one listening position but it will degrade for all others. More likely, the reproduced sound will change without definitive improvement. Hence, those concerned with the reproduction of sound have recognized the need for a system and method of modeling the complete behavior of sound reproduction devices such that conjugate responses to the sound repro duction device responses may be created. The need for a system and method employing modifiable conjugate response has also been recognized. Furthermore, the need has also been recognized for a system and method that compensates the reproduction of sound independent of the environment in which the sound is to be heard. The present invention fulfills these needs and others. SUMMARY OF THE INVENTION ss 60 65 Briefly, and in general terms, the present invention pro vides a system and method for modeling individual response characteristics of a sonic reproduction device to create a conjugate model for improving frequency, time, phase, and amplitude performance of the device and to provide improved sonic balance, sound clarity, reduced distortion and improved stereo imaging. In a first aspect, the invention relates to an apparatus for modifying an electrical audio signal for input to a sonic reproduction device characterized by a plurality of indi vidual responses. The individual responses of the device combine to define an overall response. Each individual response includes one or more of a frequency, time, phase or US 7,184,556 B1 17 provides a dead-band or band-reject capability to accom modate manufacturing tolerances from one speaker to another. Side frequency boost is still needed and double tuned resonators are best used. Other variations and simplifications can be made when adjacent behavioral modes have similar properties, as is likely with most speakers. Two or more rejection notches can share Q and amplitude settings as well as compensation boost. Combinations include two notches with three boosts, two notches with two asymmetrical boosts, three with two, etc. A single low-Q boost with a frequency halfway between two notches can be used. Three low-Q boosts with frequen 10 cies below, above, and between are a better variation. For all of these implementations, the notch depth is often great and the side frequency boost is usually small. Usually, the overall energy response to random noise averaged about the compensation region is made to be the same or slightly higher than without correction. Delayed Interference or All-Pass—A hybrid analog-digi tal CCD device can create a small, convenient tunable delay. Though performance may be poor, they can be connected or configured like the example in FIG. 16 to provide interfer ence-like behavior. The circuit can create approximate con jugate responses to wavelength related reflection and trans mission behavior from walls, tables and the insides of speaker enclosures or other parts of the transmission path or system. The circuit can be set to create an inverted comb filter or additive interference like response which would be opposite in time, phase and amplitude to subtractive inter ference loss from reflecting surfaces. The correction boosts where interference takes away. The circuit can also be adjusted to have a comb filter like response to cancel additive energy from reflections within the speaker enclo sure. Better time delay interference filters or comb filter like responses can be made from DSP processes. Both the analog and the DSP can be configured to be relevant to the physical reflection model and like other parts of the correction system, are controlled by parametric adjustments related to physical behavior. The delay interference path filter has controls relating to dimensions, surface absorption, and the 15 20 different corrections are needed from one installation to 25 30 35 40 amount of interference correction needed. With reference to FIG. 16, Ta relates to the difference between the direct path from the speaker to listener and the longer bounce path also from speaker to listener. To also relates to the out and return path between the speaker and an opposite surface inside an enclosure. A wall behind the speaker can be characterized the same way. Larger Ta gives a larger distance. RC relates to surface roughness or absorp tion at high frequencies. Larger RC product for greater loss or faster attenuation of upper frequency comb filter response and correction. The control R1 adjusts the magnitude of the response or correction. CW direction increases subtractive responses while the CCW position near the +input to the op amp gives maximum additive responses. The circuit pro duces an interference response whose amplitude decreases with frequency. This matches or simulates losses of absorp tion materials of practical speakers. Much of the irregular response from small speakers can be experimentally changed to something that appears to be more easily pro cessed by the compensation system. Usually, the delay setting to do this matches the back arrival wave relationship expected from the speaker enclosure. When it does, this one adjustable parameter equals a multitude of conventional response-leveling processes. FIG. 17 is an all pass or phase shift network. Its frequency response is flat however its output is in phase and high frequencies and out of phase at low frequencies. The circuit 18 alters transient response without changing frequency response. The variable control increases the transition fre quency as it is turned CW. This element is useful to correct group delay and other transient related responses. As previously mentioned, mechanical disturbances pro duced by speakers are often audible, yet invisible or hard to interpret from response measurements made using a fre quency sweep and microphone. Traditional compensation methods using a deep notch usually result in either a nasal sound or undesirable balance aberration. The weighted com pensated notch filter of the present invention solves this problem and yields some other advantages as well. When two drivers (woofer and tweeter) are crossed over by just a capacitor or by circuits that overlap frequencies, one or both drivers can have interference compensation without percep tual loss to the other. The same applies to different listening positions. One position having a bad response can be com pensated without compromising the sound for other listen ing positions. The correction is hidden by the weighted side energy. Since good listening positions are not compromised by the correction, a wide range of listening positions can have good sonics. This feature is particularly useful to horn-type loudspeakers for theater sound, where slightly 45 50 ss 60 65 another. By using default optimization for a class of speakers or construction features, parametric control and adjustment is simple and intuitive. Some of the most important behaviors of loudspeakers (with respect to acoustically perceptible effect) cannot be modeled or implemented from traditional methods. Such behaviors include interference and resonant coupling, as well as nonlinear consequences from such behaviors. In addition, other spatial factors related to stiffness of moving parts and high frequency de-coupling for motions away from a driving voice coil need to be considered. Any of these can create source movements, delayed energy release and phase error to binaural hearing. Often, such destructive responses can be invisible or very difficult to interpret from traditional microphone-and spectrum-analyzer calibration methods. An example of a parametrically addressed compensation for a specific physical effect is compensation for mechanical de-coupling in large full-range speakers. Such speakers are usually designed to have the entire cone move at low frequencies. At high frequencies, only the inner part of the driver is the primary active radiator. The rest of the cone is intentionally de-coupled to attenuate its nodal breakup. This design choice changes the position ofan equivalent radiation source. A linear correction to move high-frequency radiation forward in time uses complex pass filters to create band limited delay of the lower frequencies. Then, as frequency increases, a latent delay decreases thereby maintaining a phase match to the response order of the speaker. Some group delay distortions can be removed this way. A physical dimension from the speaker along with attenuation and speed-of-sound properties of the cone can yield information to specify a correction from a dedicated filter process. However, the delay effect and de-coupling frequencies can be experimentally determined to yield parameter values. If some of the speaker parts change size or attenuation prop erties, new values can be extrapolated. An additional example of a parametrically addressed compensation, which is difficult or impossible to compen sate using traditional methods, is compensation for Doppler modulation in small speakers. Computer and multi-channel sound systems require small speakers to play very loud. The resulting combination of fast, high-displacement cone motions can impart an additive or subtractive velocity to the US 7,184,556 B1 25 mass, air volume, mechanical compliance, radiating area, damping, moving mass and motor characteristics. 7. The system of claim 1 wherein the sonic reproduction device comprises a speaker and at least one of the plurality offilters comprises at least one associated adjustable param eter and the value of the parameter is derived from a standard speaker model. 8. The system of claim 1 wherein at least one of the plurality of filters has at least one associated adjustable 26 parameter and the value of the parameter is determined experimentally using standard test measurements. 9. The system of claim 1 wherein the sonic reproduction device comprises a speaker and the one parameter that modules the at least one other parameter relates to the magnet structure and voice coil of the speaker.
* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project