WO2009010056A1

WO2009010056A1 - Method of determining a class of a load connected to an amplifier output

Info

Publication number: WO2009010056A1
Application number: PCT/DK2007/050100
Authority: WO
Inventors: Klas Åke DALBJÖRN; Kent Tange; Esben Skovenborg
Original assignee: The Tc Group A/S
Priority date: 2007-07-16
Filing date: 2007-07-16
Publication date: 2009-01-22
Also published as: WO2009010069A1

Abstract

The invention relates to a method of determining a class of a load by - providing at least one classifier representing at least one class, - measuring and extracting at least one feature of a measured load, - determining the class of said measured load statistically on the basis of said at least one feature and said at least one classifier. The invention further relates to a load class determining amplifier comprising an amplifier, a data processing means and an amplifier output connected to a load, said amplifier comprising means for measuring and extracting at least one feature of said load, and said data processing means comprising means for determining the class of said load statistically on the basis of said at least on feature and at least one classifier. The invention further relates to a method of verifying if a load connected to an amplifier output corresponds to a predefined load, said method comprising the steps of - measuring and extracting at least one feature of a measured load, - determining statistically on the basis of said at least one feature and at least one classifier representing at least one class if said measured load belongs to the class of said predefined load.

Description

METHOD OF DETERMINING A CLASS OF A LOAD CONNECTED TO AN AMPLIFIER OUTPUT.

Field of the invention

The invention relates to a method of determining a class of a load according to claim 1.

Background of the invention

A certain degree of monitoring of a load coupled to an amplifier is well-known, both during live -use and in test setups.

An example of such disclosure is found in US 2006/0104453 Lee where loudspeakers in a multi-channel system are analyzed with respect to impedance to determine if they are duct- type, and if they are small or large type loudspeakers. A problem of such rule-based and "hard logic" analysis is that the resulting determination of the load are inherently relative primitive and simple as the analyzed load basically is a non-linear and complex element.

Summary of the invention

The present invention relates to a method of determining a class of a load by

- providing at least one classifier representing at least one class,

- measuring and extracting at least one feature EFC of a measured load LS; 2, 6,

- determining the class CL of said measured load statistically on the basis of said at least one feature and said at least one classifier.

Broadly, the invention relates to a class as a collection of loads that have something in common, e.g. model, band and/or driver of the connected load. The class may thus broadly be established and applied for the purpose of establishing a sort of reference to which a measured load may relate. The class(es) may thus be very broad or non- detailed if the purpose of determining the load is only to discriminate between an LF- driver, and MF-driver and a HF-driver. Likewise, the class(es) may be very narrow or detailed if the purpose is to discriminate between different specific drivers or cabinets of same loudspeaker types. In an embodiment of the invention, different degrees of broad and narrow classes are established among which the class of the measured load can be determined.

One of the significant advantages of the invention is that the determination is established statistically instead of based on a set of rules, thereby availing much more gradual determination when compared to a rule-based determination based on when the measured load falls into one interval of reference measures or not.

According to an embodiment of the invention, determination on the basis of a classifier and measured features is preferably performed on the basis of the classifier having the measured features as input.

It should be noted that a statistical method may comprise rule-based decisions or processing, as long as just a part of the processing is based on statistics. In other words, methods that are not pure rule-based, are statistical methods and lead to a statistical determination. Furthermore, it should be noted that a classifier according to the present invention may be a simple measure such as, e.g. a statistical distance measure, or it may comprise complex processing, including pre-processing for preparing statistical data or the measured and extracted features, e.g. a statistical normalization, and post-processing for analyzing and interpreting the results, e.g. a decision layer. Hence, a statistical determination according to the present invention may comprise rule-based processing, e.g. in a decision layer, or as part of the core classifier method itself.

The statistical approach of the present invention further avails modeling of classes applicable for the determination which may otherwise be difficult to establish robustly by a set of rules.

In order to establish whether a measured load belongs to or is "close/closest" to a certain class it must be possible to relate the measured load to a representation of one or more classes in the classifier. This representation may be in the forms of a statistical model of the class, e.g. a representation based on relevant parameters which may facilitate an automated determination of the class. The model may be established in numerous different ways within the scope of the invention.

Another advantage of the present invention is that the classes that are represented in the classifier need not be explicitly defined. Due to the statistical nature of the modeling in the classifier, the classes may be implicitly defined during a "training" or "model parameter estimation" phase. This training may be performed only once, based on a pre-selected set of classes. Or, the training may be carried out at any time, when either a new class is to be added to the system, or when an existing class is to be extended or its model be further substantiated.

The classifier may thus be regarded as a representation of one or more classes represented in a multi-dimensional feature space.

A classifier may for example comprise a statistical function based on a set of measured features of reference loads belonging to the same class or classes (non- parametric model) or a classifier may for example comprise a statistical model consisting of parameters optimised or estimated based on measured features of reference loads belonging to the same class or classes (parametric model).

It should be noted that a model may be applied for representation of several classes, e.g. when applying a neural network model as classifier.

A determination may be established in several different ways within the scope of the invention. One way of establishing a determination may thus e.g. be a verification where the input to the process is a request regarding whether the measured load is belonging to a certain class. The response to the request may be established in several different ways but it may simply be a yes or a no to whether the load has been determined to belong to the requested class. A response may also involve that a probability measure is established and returned to the user. Another way of establishing a determination may e.g. be to establish a number of probability measures related to different modeled classes in order to avail the user to know the most likely load class.

Several other ways of establishing the determination may be applied within the scope of the invention as long as the determination is based on a statistical approach.

It should also be noted that the invention facilitates very complex and user friendly determinations of class(es), in that the amount of information established with the determination and the amount of information or options presented to the user may range from very simple to very complex, according to the needs and interests of the user, and the complexity and extensiveness of the classifiers, possible classes, possible loudspeakers used, the loudspeaker setups tested, etc. Thus, as one of several possible examples within the scope of the invention, a determination may e.g. be established as a verification request where a load is measured and where the result of the determination is a "reply" establishing that a narrow class could not be determined for the specific loudspeaker, but that the measured loudspeaker was determined to fall within a "super-class" of subwoofers. Several other examples of such determinations may be applied within the scope of the invention.

This aspect also involves that the determination of a class of the measured load can lead to several different types of results in different embodiments of the invention. The determination result may in an embodiment merely be a "soft" determination, e.g. a probability for the measured load belonging to one or more classes, i.e. without any decision layer for considering or interpreting the probability determined. This must then be done by the user or other processing means. In a different embodiment, the determination includes a decision layer, e.g. establishing a result comprising the class to which the measured load most likely belongs, or a verification result as mentioned above.

According to the present invention a feature of a measured load may be any feature derivable from measurements, typically electrical measurements, or the measurements themselves. Features include, but are not limited to, voltage, current, corresponding values of voltage and current, impedance, characteristics derived from impedance as function of frequency, e.g. mean value, variance, maximum slope, minimum slope or an expression for 'ripple', etc., possibly divided into several frequency bands, DC resistance, resonance frequency, power consumption, transfer function, temperature, physical properties, impedance change according to applied power, temperature change according to applied power, cone excursion caused by applied power or voltage, difference of response to small signal and large signal stimuli, impulse response to a specific test signal, integrated, averaged or differentiated functions of any of the above features during a time window or in a frequency band, specific electrical or physical reaction to a specific test signal, vibration caused by a specific test signal, etc.

The measured and extracted features may be specifically related to a loudspeaker or other load for which determination of a class is desired, or they may relate to the loudspeaker or other load in combination with an additional load, e.g. a loudspeaker cable. According to an embodiment of the present invention, a class determination may be performed for a part of a combined load, e.g. a loudspeaker being part of a cable and loudspeaker combined load, as the statistical approach of the present invention facilitates determination of class even for loads being influenced by additional loads such as significantly long loudspeaker cables. In an alternative embodiment of the present invention, determination of a class of loudspeaker cable or a class of a combination of a loudspeaker cable and a loudspeaker is facilitated.

Hence, a load according to the present invention may be any load from which features may be measured and extracted as described above, and comprise a combination of loads, e.g. a cable and loudspeaker, or two or more loudspeakers coupled in parallel or series. It is noted that a load according to the present invention relates to more than loudspeakers per se, and may relate to any kind of loudspeakers, headphones, monitor loudspeakers, in-the-ear-monitors, hearing aids, transducers, megaphones, specialized loudspeakers, loudspeaker arrays, loudspeaker cables, channel separation filters, attenuators, subsequent amplifier stages, wireless audio transmitters, converters, powered microphones, probes or sensors or any other kind of loads typically or possibly coupled to amplifier outputs of any kind or any other means of establishing a signal, typically an audio signal.

As an example, in an embodiment of the present invention the determination is performed by means of classifiers based on a statistical distance measure D, and the features measured and extracted from the load are mean value and variance in several different, overlapping feature extraction frequency bands Bl, B2, ..., B7 derived from a load impedance function established by a method and/or an amplifier as described in PCT patent application No. PC17DK2007/050099 filed on 16 July 2007, hereby incorporated by reference as regards a method and amplifier for establishing an impedance function of a connected load.

When said method of determining a class of a load comprises the steps of - providing at least one classifier representing at least one class,

- measuring and extracting at least one feature EFC of a measured load LS; 2, 6 connected to an amplifier 1 ,

- determining the class CL of said measured load statistically on the basis of said at least one feature and said at least one classifier, an advantageous embodiment of the present invention is obtained.

An amplifier may typically comprise one or several amplifier outputs to which a load, typically a loudspeaker may be coupled. The invention moreover allows individual determination performed in relation to each amplifier output and the associated load and it may also facilitate analysis of one or more groups of the amplifier outputs.

The establishment of such a determination will be described more in detail below as well as a description of suitable features which may serve basis for a statistical determination, i.e. based on probabilities. When allowing determination based on probability measures more complex and detailed determination may be established than by a straightforward, purely rule-based match/ no-match analysis as the loads typically are fundamentally non-linear and therefore unsuitable for straightforward linear deduction, e.g. a determination where different parameters are considered sequentially. Such non-linear behavior may e.g. rely on thermal conditions, aging, stress, production tolerance, etc. In other words, determination according to the present invention facilitates recognition even of loads which may look very much the same when performing primitive rule -based recognition, e.g. peak detection in impedance characteristics.

It is noted that a variant of determination of loads also within the scope of the present invention is verification that loads correspond to predefined or expected loads.

Where determination of a class of a load in general comprises a set of possible classes containing several possible loads and the special class unknown, the variant of verification that a load corresponds to a certain expected load can be seen as a determination of a class of a load where the set of possible classes contains only a single class and the special class unknown, which in that case means not verified. An alternative way of deriving a simple verification from a general determination result is to compare the determined class with the expected class. This is a logical task leading to a binary result as verifications inherently do. Any degree of class determination and any result derivable from it, with any degree of information detail, and any variants such as verification, are within the scope of the present invention.

When the method is applied within an amplifier 1 and wherein said measured load LS; 2, 6 is connected electrically to said amplifier 1, an advantageous embodiment of the present invention is obtained.

The load may preferably be connected to the amplifier by conventional cabling.

When the amplifier executes the method by means of data processing means 3 according to instructions stored in memory means, an advantageous embodiment of the present invention is obtained. The method is preferably computer implemented, i.e. it is preferably implemented by data processing means implemented in a personal computer, a laptop, a digital signal processor DSP, a microcontroller or microprocessor, a field programmable gate array FPGA, an application specific integrated circuits ASIC, etc.

When the amplifier comprises a data port by means of which models may be transferred to or from the amplifier, an advantageous embodiment of the present invention is obtained.

An applicable data port may e.g. comprise a network port, e.g. an Ethernet network port, USB port, Firewire port, Infrared port, Blutooth port, etc.

When said classifier or data related with said classifier is stored in a data storage 5, an advantageous embodiment of the present invention is obtained.

When said amplifier 1 comprises said data storage 5, an advantageous embodiment of the present invention is obtained.

When said data storage 5 is comprised by a central data storage to which said amplifier 1 has irregular or continuous access, an advantageous embodiment of the present invention is obtained.

According to an embodiment of the invention, the amplifiers may be coupled in a data network, typically with a central controller or monitoring means. In such a network, the data storage containing the classifiers and other relevant data may as well be comprised by a central network unit, e.g. as part of the central controller or monitoring means, in order to ensure that all amplifiers have access to the same information. In a further embodiment, the amplifiers may have access to an extended network, e.g. the Internet, e.g. for participation in a community for classifier distribution or for remote access to classifiers. The network connection, e.g. in the case of access to the Internet, need not be continuously established but may be established on demand or according to the needs. When said method comprises a decision layer, an advantageous embodiment of the present invention is obtained.

In a preferred embodiment of the present invention, a decision layer is provided to facilitate the user in interpreting and considering the results of the determination. The decision layer may, e.g., consider established probabilities and decide the most probable option, or it may, e.g., convey to the user that all loads are correct except from a few specified loads which are problematic to determine. As noted above, the decision layer can be rule-based or statistical, and it can form part of a classifier or be a separate method step carried out after the statistical processing, and possibly only on request from the user.

When said at least one classifier comprises a parametric classifier, an advantageous embodiment of the present invention is obtained.

When said at least one classifier comprises a non-parametric classifier, an advantageous embodiment of the present invention is obtained.

When said at least one class is established by providing a set of reference features RFC of a plurality of reference loudspeaker units RLS, an advantageous embodiment of the present invention is obtained.

When at least one of said reference features RFC is provided as a function of frequency, an advantageous embodiment of the present invention is obtained.

When at least one of said features EFC is determined as a function of frequency, an advantageous embodiment of the present invention is obtained.

According to a preferred embodiment, the features are considered with regard to frequency dependency, as the extra dimension of frequency in most cases widens the data set significantly and causes otherwise similar looking data sets to reveal significant differences.

When said determination of a class of said load LS; 2, 6 is based on a probability of match between said measured and extracted features EFC of said measured load and one or several reference features RFC, an advantageous embodiment of the present invention is obtained.

When said determination of a class of said load LS; 2, 6 involves a statistical determination, an advantageous embodiment of the present invention is obtained.

When said determination of a class of said load LS; 2,6 involves a non-linear signal processing, an advantageous embodiment of the present invention is obtained.

By non-linear signal processing is facilitated more complex analysis and non-linear determination than facilitated by only linear processing means.

When the determination is performed automatically, an advantageous embodiment of the present invention is obtained.

When the determination is performed semi-automatically, an advantageous embodiment of the present invention is obtained.

When the determination is performed during a calibration phase, an advantageous embodiment of the present invention is obtained.

A calibration phase is regarded as the initial phase when a load has been coupled to an amplifier and where the initial tests are performed.

When the determination is performed during a verification phase, an advantageous embodiment of the present invention is obtained. According to a preferred embodiment the method of determination is performed during a verification phase where it is verified if the connected load for each channel in the system is of the make and model that the user has predefined in his system configuration, i.e. that the class of each connected load corresponds to the predefined class. This is a convenient way to detect that everything has been connected as intended in a large system (often containing several hundred amplifier channels).

When said determination of a class of said load LS; 2,6 involves indication of one or more of the make, model, band, driver and/or number of parallel coupled loudspeakers of the measured load, an advantageous embodiment of the present invention is obtained.

The resulting determination of class may be more or less detailed depending on the intended use of the result and also depending on how many different loads which are actually applicable.

An example of a make and model indication is, e.g., JBL Vertec VT 4889. An example of a detailed make, model and band indication for a channel is, e.g., the MF band of a JBL Vertec VT4889, as this 3-way loudspeaker has separate inputs for each band. Such detailed information can be used for looking up further characteristics in the model's data sheet, preferably digitalized data sheets for automatic use by the amplifier.

When said determination of a class of said load LS; 2,6 involves indication of the type of the measured load, an advantageous embodiment of the present invention is obtained.

An example of a type indication is, e.g., that the connected load belongs to the class of one or several woofers in bass reflex type cabinet(s). Such information can be used for looking up generic characteristics for that type of loudspeaker, e.g. for further use by the amplifier. When said determination of a class of said load LS; 2,6 involves indication of possible classes to a user and where the possible classes are associated with indication of calculated probabilities, an advantageous embodiment of the present invention is obtained.

By listing of calculated probabilities a user, e.g. a sound engineer, may manually complete the determination on the basis of knowledge of what loads the user expected to be connected to the amplifier. Optional or probable determinations may be displayed to a user and the user may then use his knowledge and expectations of the system to decide the established class or load.

When said determination of a class of said load LS; 2,6 involves indication of the load being a predefined load, or the probability of the load being a predefined load, an advantageous embodiment of the present invention is obtained.

According to this embodiment, the user, e.g. a sound engineer, makes the intended system configuration, i.e. which loudspeaker is intended on which channel, available to the system, and the determination method involves a decision layer verifying this configuration for each channel, or establishing probabilities of correct connections for each channel.

When said features EFC are extracted in at least two separate or overlapping feature extraction bands Bl, B2, ..., B7, an advantageous embodiment of the present invention is obtained.

According to an advantageous embodiment of the invention feature extraction may be performed in different bands with respect to frequency in order to establish a number of features sufficient to allow distinguishing between loads which are behaving relatively equal. When said features EFC are extracted in at least three overlapping feature extraction bands Bl, B2, ..., B7, an advantageous embodiment of the present invention is obtained.

According to an advantageous embodiment of the invention feature extraction may be performed in different overlapping bands with respect to frequency in order to establish a number of features sufficient to allow distinguishing between loads which are performing relatively equal.

When said features EFC are extracted in at least seven overlapping feature extraction bands Bl, B2, ..., B7, an advantageous embodiment of the present invention is obtained.

According to a preferred embodiment of the invention, the features are extracted in relatively many bands with respect to frequency in order to improve the robustness of the class determination.

When the determination of a class of said measured load is applied for deriving of information associated with said class CL, an advantageous embodiment of the present invention is obtained.

According to a preferred embodiment of the invention, determination of the coupled load may be applied for the purpose of establishing information which is associated with the determined class and application of this information in connection with the normal use of the amplifier when coupled to the load. Thus, such information may facilitate a live monitoring of the temperature of the voice coil, determination of aging, etc.

Moreover such normal use may e.g. include automatic or at least semi-automatic determination of short-circuits, defect cables, incorrectly connected loads, etc. When said at least one feature EFC comprises at least one electrical measurement or is derived from at least one electrical measurement, an advantageous embodiment of the present invention is obtained.

When said at least one feature EFC comprises impedance of the load as function of frequency, variance and/or mean value of the impedance of the load, resonance frequency of the load, DC resistance of the load, etc., an advantageous embodiment of the present invention is obtained.

It should be noted that e.g. mean impedance in two different feature extraction bands Bl, B2, ..., B7 - overlapping or not - may be regarded and applied separately as two different features. Hence, e.g. 14 individual features are determined by determining 2 features in 7 feature extraction bands.

When said measured and extracted features are compensated for a cable component before said determination of class is performed on the basis of said features, an advantageous embodiment of the present invention is obtained.

According to a preferred embodiment of the present invention, information about the impedance or other significant features of a loudspeaker cable or other cable is established in order to be able to neglect or subtract this information during the determination of class of the load or other use of the features measured and extracted from the load.

When at least two features EFC are established and analyzed in different feature extraction bands Bl, B2, ..., B7, overlapping or non-overlapping, an advantageous embodiment of the present invention is obtained.

The present invention further relates to a load class determining amplifier, comprising an amplifier 1, a data processing means 3 and an amplifier output AO; 4 connected to a load LS; 2, 6, said amplifier 1 comprising means for measuring and extracting at least one feature EFC of said load, and said data processing means 3 comprising means for determining the class CL of said load statistically on the basis of said at least one feature and at least one classifier.

When said load class determining amplifier comprises means for carrying out a method of determining a class of a load according to any of the above, an advantageous embodiment of the present invention is obtained.

The present invention further relates to a method of verifying if a load LS; 2, 6 connected to an amplifier output AO; 4 corresponds to a predefined load, said method comprising the steps of

- determining statistically on the basis of said at least one feature EFC and at least one classifier representing at least one class if said measured load belongs to the class of said predefined load.

When said method of verifying if a load corresponds to a predefined load comprises any of the features of the above described method of determining a class of a load, an advantageous embodiment of the present invention is obtained.

The present invention further relates to a load verification amplifier comprising data processing means 3 for carrying out a method of verifying if a load corresponds to a predefined load according to the above.

The present invention further relates to a system comprising an amplifier 1 according to any of the above, and at least one load 2, 6.

The present invention further relates to a use of a method according to any of the above.

The present invention further relates to a method of distributing classifiers representing load classes or data related to said classifiers, the method comprising providing a central data storage CD comprising a data port 130 and enabling at least two amplifiers 1 or users of amplifiers 1 to exchange said classifiers or said data with said central data storage CD.

The present invention further relates to a community for distribution of classifiers representing load classes or data related to said classifiers, said community comprising at least two amplifiers 1 or users of amplifiers 1 and a central data storage CD, said amplifiers 1 and central data storage CD comprising data ports 130 for facilitating exchange of said classifiers or said data.

The determination of a class of a load according to the present invention relies on the availability of classifiers or data from which classifiers can be established, i.e. data from preferably several reference loads belonging to the same class. As mentioned above, one way to establish a classifier is from measurements, e.g. impedance function measurements, for a number of loudspeakers belonging to the same class. Such measurements and classifier establishment can evidently be performed by the loudspeaker manufacturers. It may, however, be difficult to get the loudspeaker manufacturers to establish classifiers for their loudspeakers, including the loudspeakers already on the market and possibly discontinued, and it may therefore be insufficient to rely on the loudspeaker manufacturers to establish a huge classifier database for use in the amplifiers. Moreover, it may be beneficial to establish classes or classifiers representing super-classes, alternative loads, e.g. loudspeaker cables, loudspeakers damaged or worn in distinctive ways, both as sub-classes of type- classes, e.g. class of tweeters with a bulged cone, and as sub-classes of narrow model-classes, e.g. that specific tweeter model with a bulged cone, etc. Hence, even though the loudspeaker manufacturers may establish classifiers for their loudspeakers, it may be advantageous to facilitate a classifier establishment and distribution not limited to the loudspeaker manufacturers. Moreover, a means for distributing new classifiers should be established in order for users to be able to keep their amplifier load databases up to date when acquiring new loudspeakers.

According to the present invention, an advantageous embodiment is to download classifiers to the amplifiers 1 from other amplifiers 1 or from the central data storage CD, or from a classifier provider CP. Thereby is facilitated updating the local loudspeaker profiles represented by classifiers, e.g. as new loudspeakers are acquired.

According to the present invention, an advantageous embodiment is to facilitate the amplifiers 1 to upload measured impedance functions or other data they have determined. As the classifiers get more robust and certain with larger data sets, and as the amplifiers are able to measure an impedance function or other data necessary to classify a load, in fact all the amplifiers are able to cooperate in establishing data sets for the classifiers. In other words, each time a load is determined by an amplifier, and possibly verified by a user, the measured data can be uploaded to, e.g., the central data storage or a classifier provider. The data may be accompanied by related, measured or manually input data, e.g. regarding temperature, connection, etc. Also new classifiers or data related to loads not yet represented by a classifier may be uploaded. In that case the user should preferably add data about the desired or suggested class, or any other relevant data such as, e.g., degree of wearing, any damages, type of use, etc. In the case that classifiers are provided to other community users on the basis of few measurements e.g. from other users, the classifiers may be associated with a robustness score or other indicator representing to what degree the output of the classifier can be trusted.

In a community as described above, the classifiers can be free to download, or download can be subject to a charge. In a preferred embodiment the charge and availability can be decided by the central data storage or provider. In a preferred embodiment a user contributing by uploading data or a classifier receives a compensation in the form of e.g. money, virtual money for use in the community, or the right to download a number of classifiers for free. In a preferred embodiment the classifiers can be discussed, scored, suggested changed, etc., by the community users. In a preferred embodiment the community may further facilitate distribution of amplifier related data other than load classifiers, e.g. amplifier settings, etc. An advantage of the present invention is that the measurements are performed locally, i.e. not by a central instance, thereby relieving the loudspeaker manufacturers, the amplifier manufacturers or dedicated companies from performing them. Moreover, and even more important and beneficial, the measurements are made in real life situations, by real life amplifiers with real life speakers with natural wear and characteristics. Thereby, the data sets and classifiers established will possibly better handle classification in the actual live situations for the users downloading classifiers established this way.

The establishment of classifiers on the basis of (locally performed) measurements should preferably be performed centrally, e.g. by a central data storage, a classifier provider, the amplifier manufacturer or a dedicated classifier company in order to ensure quality of the established classifiers, an in order to avoid errors and maintain a structured and user friendly classifier hierarchy. In alternative embodiments, a user may be able to establish a classifier by means of several measurements on different loads of same class. In an alternative embodiment, such a locally established classifier may advantageously be used locally by the user establishing it, and may, if shared with the community, be marked as 'homemade'.

If the establishment of classifiers relies on extraction of features, e.g. from measured impedance functions, e.g. for use in a statistical distance classifier, the features may be extracted in the amplifiers before upload of data to the central data storage, or they may be extracted by the central data storage when establishing the classifier.

The drawings

The invention will now be described with reference to the drawings of which

fig. IA and IB illustrate an amplifier according to an embodiment of the invention, fig. 2 illustrates a principle of determination according to a preferred embodiment of the invention, fig. 3 illustrates an impedance characteristic for a 3-way loudspeaker, fig. 4 illustrates an amplifier according to an embodiment of the invention, fig. 5 illustrates an impedance characteristic for 1 to 4 loudspeakers coupled in parallel to one amplifier output, fig. 6 illustrates a block diagram of the different steps in a process according to an embodiment of the invention of classifying or verifying a loudspeaker coupled to an amplifier output, fig. 7 illustrates a principle of overlapping feature extraction bands according to an embodiment of the present invention, fig. 8 illustrates an example of a data set for a statistical distance classifier, fig. 9 and 10 illustrate specific examples using a method according to embodiments of the invention, fig. HA and HB illustrate flow diagrams of automatic classification methods according to embodiments of the present invention, fig. 12A and 12B illustrate flow diagrams of automatic verification methods according to the present invention, and fig. 13A and 13B illustrate embodiments of classifier distribution communities according to an embodiment of the present invention.

Detailed description

Fig. IA illustrates an amplifier facilitating determination of a class of a coupled load according to an embodiment of the invention.

The amplifier is connected to a load 2 via an amplifier output 4 by means of a cable. The cable may comprise one or several conductors, typically two conductors.

Further loads 6 may be coupled to the amplifier both by means of separate dedicated amplifier outputs as illustrated in fig. IA or e.g. by parallel coupling to one amplifier output, as illustrated in fig. IB.

The amplifier moreover comprises a signal processor 3 by means of which a load coupled to the amplifier output 4 may be analyzed. The signal processor 3 comprises or is associated to a data storage 5. In particular the signal processor should facilitate feature extraction applicable for automatic or semi-automatic determination of a class of the load.

The amplifier 1 may be a stand-alone amplifier or it may be distributed in two or further units. The amplifier may be any kind of amplifier, including analog amplifiers of any kind, e.g. class B, class AB, class G, class H, etc., switching amplifiers of any kind, e.g. class D, etc., or a hybrid amplifier like the so-called tracked class D amplifier, described in more detail in U.S. patent No. 5,200,711, hereby incorporated by reference.

The functioning of the above disclosed embodiment of the invention will be further disclosed below.

Fig. 2 illustrates a principle of determination of a class of a load according to a preferred embodiment of the invention. As just one of several examples within the scope of the invention, the determination will be explained with reference to the exemplary embodiments of fig. IA and IB. Other hardware setups may be applied within the scope of the invention.

The determination involves a reference data base DB which e.g. may be stored in the data storage 5 of fig. IA and IB. The data base comprises a number of classifiers or data related to classifiers or representing classes represented by classifiers. Such data may, e.g., comprise reference features RFC. Each reference feature corresponds and describes relevant features of loads which may be coupled to the amplifier output.

In other words, features EFC of the load connected to the amplifier output 4 in figure IA and IB is measured and extracted, preferably as a function of frequency, during a calibration phase or during use - live use.

The measured features EFC of the coupled load are then input to the classifiers related to the data in the database DB, e.g. a set of reference features RFC for the purpose of determining a class to which the coupled load belongs.

As a specific example, the set of reference features RFC may comprise features of a number of different loads which have been measured and analyzed previously and the resulting features are then stored in relation to the amplifier system. The actual measured coupled load may then be subjected to one or more of the classifiers of the data base and therefore serve as a basis for an automatic or semi-automatic determination of the class of the coupled load.

The establishment of such a determination will be described more in detail below as well as a description of suitable features which may serve basis for a statistical class determination, e.g. based on probabilities. When allowing determination based on statistical or probability measures more complex and detailed determination may be established than by a straightforward, purely rule-based match/ no-match analysis as the loads typically are fundamentally non-linear and therefore unsuitable for straightforward linear deduction, e.g. a determination where different parameters are considered sequentially. Such non-linear behavior may e.g. rely on thermal conditions, aging, stress, production tolerance, etc. In other words, determination of class according to the present invention facilitates recognition even of loads which may look very much the same when performing primitive rule-based recognition, e.g. peak detection in impedance characteristics.

Thus specific examples of runtime monitoring of a coupled load may be established by using the output voltage and current of an amplifier generated by the music signal through the same output, e.g. estimation of the voice coil and magnet temperature when music is playing and when there is silence, detection of changes in loudspeaker setup, for example one loudspeaker is disconnected, or detection of open/short circuit at output.

Fig 3 illustrates examples of impedances at the vertical axis as function of frequencies at the horizontal axis of 3 different driver units aimed at handling different frequency bands, e.g. as comprised by a 3 way loudspeaker. The curve 31 with the high peak at about 65 Hz illustrates the impedance characteristic of an LF driver for reproducing audio at low frequencies with respect to the audio band, the curve 32 having two small peaks below 200 Hz illustrates the impedance characteristic for an MF driver for reproducing audio at medium frequencies with respect to the audio band, and the curve 33 being relatively flat below 200 Hz illustrates the impedance characteristic for an HF driver for reproducing audio at high frequencies with respect to the audio band. Evidently, it may be possible to some degree to classify a driver according to its type, e.g. LF, MF or HF, on the basis of the location of peaks and the number of peaks, as indicated above by a rule-based analysis. It is, however, very difficult and insecure, if not impossible, by means of pure rule-based methods to distinguish a specific driver model from tens or hundreds of drivers of the same type. As mentioned above, the present invention facilitates using statistical methods based on probability or statistical models for classifying loads, which enables more detailed determination of class, e.g. regarding subtypes or driver model, or in advanced embodiments possibly even identification of unique loudspeakers from among other loudspeakers of same type and model. In general, the determination of a class of a load need not result in a certain determined specific load class. The result may very well need further consideration or interpretation, e.g. by a user in a semi-automatic approach. Such result can e.g. be a list of probable classes with the corresponding probabilities or uncertainties mentioned. However, the method may in preferred embodiments comprise a decision layer for performing at least a part of the considerations or interpretation otherwise required from the user. In the following, the terms classification and verification are used for different kinds of results established by such a decision layer, as indicated here. The term classification is used with embodiments where the amplifier or computer connected to the amplifier is adapted to classify the actual measured load as belonging to a certain load class, e.g. type, model, driver, etc. on the basis of classifiers representing known load classes from a database containing a range of load classes or classifiers representing classes. The term verification is used with embodiments where the amplifier or computer connected to the amplifier is adapted to verify if the actual measured load with a certain, typically predefined probability belongs to a predefined load class to determine that the actually coupled load corresponds to a predefined or expected load.

A preferred embodiment of a load classification amplifier is an amplifier which is able to automatically classify all connected loads and submit a resulting, actual system configuration plan to the user, e.g. by means of a display, a printer or electronic communication. This automatic classification may advantageously be carried out when the setting up of amplifiers and loudspeakers is completed, for example in order to enable the user to easily spot any incorrect connections, e.g. LF drivers connected to subwoofer outputs. The classification may be initiated in any suitable way, e.g. by the user pressing a button or automatically each time a loudspeaker is connected or disconnected to show an up-to-date connection status, or by a central network controller submitting a classification request to all connected amplifiers, etc. A preferred embodiment of a load verification amplifier is an amplifier which is able to receive information about expected load connections, e.g. by a user uploading a complete or partial system configuration plan to the amplifier or a computer connected to the amplifier. Using the verification method the amplifier use this information to select classifiers on the basis of which the actual measured features from the amplifier outputs are classified and submits a verification result to the user, e.g. comprising which connections correspond to the expected, and which do not.

Evidently, an advanced embodiment of an amplifier or system according to the present invention may be enabled to carry out both classification and verification according to the task at hand, and probably even determination without the decision layer, e.g. for use in other processing applications, or for full control by the user. As the difference between decision layer classification and verification methods are more related to the way the user interacts with the system, than the measurements and calculations carried out, a preferred embodiment comprises measuring and processing means with a decision layer adapted for both classification and verification, and the user interface and high level algorithms are exchangeable or selectable by simple software or hardware updates, or merely options at a main menu.

It is important to mention that there is not only one method to set up the system. For small systems one amplifier may be enough to control the entire system and handle all user interaction, where in larger systems multiple amplifiers and computers can be connected to each other in a data network. Each amplifier or computer can then store a database or part of a database or a central server can store the database or at least part of the database and the user may interact with all amplifiers by means of a central user interface.

Fig. 4 illustrates a verification algorithm according to a preferred embodiment of the invention. In a user interface the user starts by selecting one or more loads 41. The user interface may be connected to a data storage 42 e.g. a database, thereby allowing the user to choose a specific loudspeaker or specific loudspeakers in the database. Alternatively, if the system does not know the expected load, the user may input the reference features RFC or classifier necessary for the method to be able to verify if the connected load belongs to an expected class not present in the database. In any case, the act of inputting may comprise any suitable method e.g. having the user browsing through available loudspeakers or loads on a small display on the amplifier, or having the user designing the full system configuration plan on a computer, e.g. a laptop, and connecting this to a network of amplifiers that thereby automatically receive relevant data according to the system configuration plan.

When the amplifier knows the expected load for one or more output channels, i.e. actually the expected classes, a step of measurement 43 is performed by measuring characteristics of the loudspeakers coupled to the amplifier. The measurement 43 can e.g. be carried out by playing a number of frequency sweeps to each or at least one of the amplifier outputs and simultaneously record corresponding estimations of voltage and current signals at the amplifier output.

The result of the measurement step 43 is used for performing impedance calculation 44 of one or more of the loads at the measured channels. When the voltage and current values are estimated or measured it is possible to do a complex impedance calculation e.g. by a multirate FFT algorithm to obtain a frequency dependent impedance characteristic for each of the tested amplifier outputs.

On the basis of the calculated impedance characteristic, different kinds of load analysis 45 can be carried out. Apart from the verification itself, described in more detail below, the load analysis also preferably comprises creating a reference 46 for use during the live performance situation for which the amplifier-loudspeaker setup is intended. By storing the initial measured impedance characteristic of the connected loads as a reference characteristics, a great advantage for the subsequent live monitoring of loads coupled to the amplifiers output is obtained. A reference impedance characteristic measured at the actual loudspeaker makes it possible to monitor if the impedance characteristic of a load is changing e.g. because of high temperature in the load or because of wear, or if a load is beginning to deteriorate or is subjected to physical damage.

Another advantageous, possible use of the calculated impedance characteristics is the estimation of the number of loudspeakers coupled in parallel 47 to a certain amplifier output, e.g. as illustrated in fig. IB. This estimation can e.g. be made by using the imaginary part of the impedance function, the real part of the impedance function or absolute value of the impedance. Examples of impedance characteristics of different numbers of equal loudspeakers coupled in parallel are illustrated in the diagram in fig. 5, comprising frequency at the horizontal axis and impedance at the vertical axis. The first curve 51 illustrates the impedance characteristic of one loudspeaker, the second curve 52 from above illustrates the impedance characteristics of two equal loudspeakers coupled in parallel to one amplifier output, and the third and fourth curves 53 and 54 from above illustrates the impedance characteristics of respectively three and four, equal loudspeakers coupled in parallel to one amplifier output. As evident from the curves the general shape of the impedance characteristic is preserved, but differently offset and scaled for different numbers of loads coupled in parallel. This quality enables the determination of the class of loads even when more loads are coupled in parallel, and it enables the estimation of the actual number of loudspeakers when first their classes have been determined and further information, e.g. the impedance characteristic for a single loudspeaker of that type, may thus be known.

By using the imaginary part of the impedance function the estimate of the parallel connections is independent of temperature. This technique may preferably be applied when dealing with subwoofers, LF and some MF drivers.

As some drivers have only a small imaginary part in the impedance, automatic determination must rely on the real or absolute value of the impedance, thereby resulting in the determination being temperature dependent. This is the case with some MF and most HF drivers. Turning back to fig. 4, the calculated impedance characteristics are further verified in the verify load step 48. An embodiment of the verification algorithm itself is described in more detail below with regard to fig. 6 - 8. The result of the verification, and possibly also estimation of number of loads coupled in parallel, is concluded by an output to the user in an output step 49. The output may be a simple confirmation of which loads correspond to the expected loads selected by the user in the first step 41, or it may be more advanced and for example indicate suggestions for the loads which could not be verified, in the line of "The load could be the correct one, but seems to be damaged", "Warning: the load seems to be of an incorrect type and damage to the load or amplifier may occur" or "The load is not the expected one, but seems to be of a corresponding type, and can probably be used with corresponding results".

Fig. 6 illustrates in details a preferred way to use a calculated impedance characteristic of an actual load for verifying that it is the correct load with respect to a predefined load, e.g. represented by a class on the basis of its known impedance characteristic or other features or statistical models.

The calculated impedance characteristic is first normalized in step 61 to establish an impedance characteristic that depends less on e.g. cable length and cable impedance, the temperature in the loudspeaker and the number of loudspeakers connected in parallel. It should be noted that the normalization with respect to e.g. variance is considered a statistical operation, and a method comprising normalization is thus considered a statistical method. Furthermore, it should be noted that normalization may be implemented as a separate pre-processing step as illustrated in fig. 6, or it may be implemented in the classifiers as part of the statistical model. In any case, a determination method comprising normalization with respect to e.g. variance is according to the present invention considered a statistical determination.

In a preferred embodiment the user may further have been requested or facilitated to input information about the cable, e.g. regarding length, cross section and resistivity, as long cables may influence the combined cable and loudspeaker impedance significantly. A loudspeaker cable of 40 meters may thus easily apply a resistance of 1 Ω (Ohm). With such data, the processing means may more accurately neglect the cable impedance from the determination. In an alternative embodiment, the temperature in the load is estimated on the basis of the calculated impedance function after the load class has been determined, and the user is asked if the temperature is probable. If not, the user is asked to input the more probable temperature from which the expected load impedance function can be calculated. The difference between the measured impedance function and the expected impedance function is then considered the cable component of the impedance, and neglected or subtracted in subsequent impedance calculations and consideration, e.g. live monitoring of temperature. In yet an alternative embodiment, the amplifier may be adapted to allow determination of the class of the cable, i.e. by performing the method of determination of the class of a load, wherein the load is the cable, e.g. a cable short circuited at the loudspeaker end, or applied with a special short circuiting plug or plug with a predetermined impedance response. To this purpose, the database should comprise classifiers representing cable classes as well as loudspeaker classes. In a simple embodiment, merely the resistance of the cable is used for compensation, and a simple impedance measurement is sufficient in order to establish cable component information.

The normalized impedance characteristic is subject to feature extraction 62 in order to establish a discrete data material preserving the characteristics of the calculated impedance function, and on which probability calculations or other statistical acts performed by classifiers can be made. In an advanced embodiment using very complex classifiers, the feature extraction and normalization as mentioned above, may be implemented as part of the classifiers instead of carried out as separate preprocessing steps. In any case, the feature extraction may according to the present invention be considered part of the statistical method. The feature extraction is preferably performed in several bands with respect to frequency of the impedance function, which is therefore preferably split into several, e.g. seven, ten, etc., different overlapping feature extraction frequency bands (Bl, B2, ..., B7), e.g. according to the weighing distribution illustrated in fig 7. The distribution in fig. 7 illustrates with frequency at the horizontal axis and a weighing factor on the vertical axis that each feature extraction band (Bl, B2, ..., B7) overlaps the one adjacent band to each side, but with decreasing weight. Thereby any distinct frequency in the audio band is in total weighed the same, either by being present and highly weighed in only one feature extraction band, or by being present and less weighed in two feature extraction bands. It is noted that any distribution, over-lapping or not, differently weighed or not, is within the scope of the present invention. Compared to a non-overlapping distribution, the overlapping distribution illustrated in figure 7 enables, however, much better detection and comparison of curve characteristics, e.g. peaks, present at the border between two feature extraction bands and avoids a characteristic, e.g. a peak, not being recognized as significant for the comparison because the actual measurement has put it in a different feature extraction band than in the reference curve in the database.

The feature extraction 62 in a preferred embodiment comprises extracting features EFC from the calculated impedance function of the actual load. Such features may in a preferred embodiment comprise e.g. mean value and variance for each of the feature extraction bands (Bl, B2, ..., B7). That is, in each feature extraction band is determined a mean impedance and the variance of the measure impedance function. In the example with 7 feature extraction bands are thereby calculated 7 mean impedance values and 7 variances. As these may in principle be considered individually and independently by the classifier, they are considered as distinct features, and the example thus leads to 14 distinct features EFC which can be input to the classifier.

One method of establishing classifiers representing loudspeaker classes is to extract reference features RFC of the reference loudspeakers beforehand in the same way as the features are extracted from the unknown loudspeaker during the determination. In a preferred embodiment, the reference features RFC are calculated or obtained beforehand, and stored in the database or formalized into statistical models in classifiers for easy lookup and subjection to the features extracted from the actual load. In an alternative embodiment, the features are extracted from both the actual load impedance function and stored reference load impedance functions at runtime. This may be beneficial if changes in the way features are extracted or classifiers established may occur in subsequent software updates or added improvements, but on the other hand the processing gets much heavier if the features are to be extracted for several reference loads at runtime. In an embodiment, the user may expect a load that is not present in the database. In such cases, the user may input a suitable classifier, e.g. by providing a reference impedance curve and let the feature extraction algorithm extract reference features and store them in the database as a classifier, or the load may have been delivered with a set of data comprising pre- calculated reference features or other data sufficient to establish a suitable classifier.

When both calculated actual features EFC and classifiers, e.g. comprising reference features RFC are at hand, a statistical method 63 can be performed in order to establish the probability or logical response of the actual load belonging to the class of the expected load. One of several applicable statistical methods within the scope of the invention is determination on the basis of calculation of a statistical distance measure indicating the similarity of an unknown sample set to a known one. One such suitable statistical distance D can be defined as

which is determined on the basis of mean value μ = {μ_λ ,μ₂, μ₃ ,..., μ_p ),

covariance matrix Σ = E[(X - E[x])(X - E[x])^τ J , and multidimensional feature vector x - [X₁ , X₁ , x₃ , ... , x_p J .

In the expression of the statistical distance above, μ and Σ constitute a (simple) statistical model, representing a class, and the data vector x consist of the features extracted on the basis of a measured load.

The statistical distance defined above is a scalar (number) that indicates how far from a modeled data set a given data vector is, i.e., how different is the data vector from the class defined by the data set. Smaller distances (i.e., smaller values for D) indicate that the data vector is likely to belong to the modeled class, and large distances indicate that the data vector is unlikely to belong to the class.

Fig. 8 illustrates a part of a data set, i.e. reference features RFC, usable by a classifier defined as described above, i.e. a statistical distance classifier, and a part of the data vector, i.e. features EFC, obtained by extracting features from the unknown load. The horizontal axis counts features, and fig. 8 shows 6 (1, 2, ..., 5, ή) features of an example data set and data vector with respect to a vertical axis of a suitable scale. Each feature may, according to the above mentioned preferred embodiment represent either mean impedance or variance in a certain frequency band, and in the case of the above-mentioned preferred embodiment, there would thus be 14 features along the horizontal axis. The data set, i.e. the reference features, are preferably established by measuring, in this example, impedance functions of several loudspeakers known to belong to the same class, e.g. several subwoofers if the classifier is for coarse graduation only, or e.g. LF drivers of several 3-way loudspeakers of the same model if the classifier is for narrow, model-wise graduation. From the several measured impedance functions are extracted, in the present example, mean impedance and variance in 7 feature extraction bands, leading to 14 distinct features from each reference loudspeaker. From this population is derived a mean reference feature for each feature, and a standard deviation reference feature for each feature. Thus, a data set is established that reflects the mean and standard deviation of each of the 14 reference features among the population of same-class reference loudspeakers. In fig. 8 is shown as an example measured reference features 81 (the circles) of a single reference loudspeaker. To represent the entire reference loudspeaker population are shown mean reference features 82 (horizontal line in middle of box) and standard deviation reference features 83 (difference between top and bottom of box). These are the reference features used by the statistical distance classifier when considering an input from an unknown load. If for example feature No. 1 in fig. 8 corresponds to mean impedance in a first feature extraction band, the reference features are mean value 82 of the mean impedances, and standard deviation 83 of the mean impedances, both of the first feature extraction band among the reference loudspeakers. If for example feature No. 2 in fig. 8 corresponds to variance in a first feature extraction band, the reference features are mean value of the variances, and standard deviation of the variances, both of the first feature extraction band among the reference loudspeakers. Features No. 3 and 4 may then, e.g., correspond to mean and standard deviation of mean impedance and mean and standard deviation of variances, respectively, in a second feature extraction band among the reference loudspeakers. In fig. 8 is further shown examples of features 84 (crosses) extracted from an unknown load. The statistical distance measure is a scalar that reflects how likely the load from which the features 84 are extracted belongs to the class that is described by the mean 82 and standard deviation 83 values. In the example in fig. 8 the unknown load represented by crosses 84 does probably not belong to the class described by the data set 82, 83, as the difference for most features, except No. 4, is great compared to the loudspeaker represented by circles 81 that per definition belongs to the class. It is noted, however, that fig. 8 only shows 6 features out of the, for example, 14 features.

Last step in the preferred verification method illustrated in fig. 6 is a decision layer 64 wherein the result of the statistical method, e.g. the calculated statistical distance D is considered for determining if the actual load belongs to the class of the expected load, or at least the probability for the actual load belonging to the class of the expected load. In a preferred embodiment utilizing the statistical distance measure method described above, a threshold distance of, e.g., 7 or 10 is predefined as the critical distance where a load is said not to belong to the expected class if the distance is greater. In an alternative embodiment, or in a classification method where no expected load is presented by the user, the reference load class with the least distance to the actual load features may be indicated as the corresponding load class, or a further distance threshold criterion may be applied to take care of unknown or damaged loads.

Several other statistical and probability processing methods may be applied within the scope of the invention as long as the method enables determination of a tested load on the basis of several features extracted from the tested load coupled to the amplifier.

For example, the following methods may be applied: • Bayesian classifier, e.g. based on a Gaussian Mixture Model of the classes

• Neural network classifier, based on a Multi-Layer Perceptron model

• Support Vector Machine classifier

Although quite different, these classifiers have some common properties, relevant for the present invention. These classifiers are all -

• Statistical, in the sense that the definition of the class or classes is done implicitly, based on a set of samples (measured load features) from each class, and in the sense that no 'rules' are used to define the classes

• Robust, in the sense that the models can "generalize" from the training data, and are robust against "noise" or variations in the actual load-measurements

• Suitable for modeling multiple classes and for modeling just a single class (like the statistical distance measure does)

These classifiers and related statistical models are widely published and are readily described in textbooks on statistical pattern recognition, such as the following, hereby incorporated by reference with regard to description of classifiers, models and statistical methods suitable for use in the present invention:

• R. Duda, P. Hart, D. Stork. "Pattern Classification", (2nd ed.), Wiley, 2001.

• T. Hastie, R. Tibshurani, J.H. Friedman, "The Elements of Statistical Learning: Data Mining, Inference, and Prediction" , Spinger, 2001.

• C. Bishop. "Pattern Recognition and Machine Learning". Springer, 2006.

• N. Cristianini, J. Shawe-Taylor, "An Introduction to Support Vector Machines", Cambridge Univ. Press, 2000.

• K. Fukunaga, "Introduction to Statistical Pattern Recognition" , Academic Press, 1990. The determination of class, with or without the decision layer, may be automatic or semi-automatic in the sense that the established probabilities may be applied directly for final determination of the type, model, make, etc. which has been coupled to the amplifier.

Such automatic determination of class may e.g. be possible if the number of load classes among which determination can be made, i.e. the classifiers represented in the reference data base of the amplifier system, are relatively low and if the load classes represented in the data base are relative easily distinguished from one another. A problem related to such a setup may of course under some circumstances result in that the determination results in: not known - not classified. It should, however, be noted that the number of features extracted from the coupled load may increase the possibility that loudspeakers looking much the same when analyzing according to conventional methods may actually be recognized. An example of this situation is given in fig. 9.

A semi automatic approach may also be that a user is presented with a number of probable matches, e.g. when using the statistical distance measure method any loads with distances less than, e.g. 20, and where the matches moreover optionally but advantageously are also associated with a probability measure, by means of which a user may deduce the probable connected load. A typical experienced user knowing the amplifiers and loudspeakers that are available to him and the differences thereof, combined with a semi-automatic system which list a few probable loudspeakers for each channel if any doubt exists, may prove very advantageous as the method of the present invention provides the user with an overview and limited range of possibilities, from which the experienced user can typically easily deduct the correct answers, and still with significant advantages compared with having the user walking from loudspeaker to loudspeaker while having a colleague at the mixer table directing audio to each channel in turn to check the connections.

In fig. 9 and 10 specific examples are given using the above mentioned method for the purpose of determining a coupled load, i.e. for classification purposes. In the specific embodiment of fig. 9, a Vertec4889 LF load is coupled to an amplifier, e.g. the amplifier of fig. IA or IB. The determination of the class of the coupled load is in this example performed by establishment of the statistical distances D shown on the vertical axis, which are calculated on the basis of 10 classifiers on the horizontal axis in the reference database. The actually coupled Vertec4889LF belongs to reference class number 7 out of 10 classes in the database illustrated in figure 9.

The bar graph shows that class number 7 clearly has the smallest distance to the measured load. According to one embodiment of the invention, the distance may typically be between 1 and 8 for a correct load, i.e. the threshold for dismissing a load as not belonging to the class of the reference is 8. It is, moreover, noted that load classes number 7 and 10 in the database actually have very familiar looking impedance curves, but that the classifier has no problem separating them in this case. Thus, a conventional rule-based approach would have resulted in that the coupled load could not be matched to any of the reference loudspeakers characteristics, or at least not be able to distinguish between loads 7 and 10, whereas the method of the present invention of applying determination of class by statistical means results in a relatively distinct recognition.

In fig. 10, a further load is tested. In this example a load unknown to the reference database is measured, and the ten statistical distances D are calculated. The specific loudspeaker used is an Adamson Spektrix MF. All the ten distances are larger than 30 as no such loudspeaker had been represented by classes in any of the classifiers in the system.

Fig. HA illustrates a flow diagram of the principle of automatic classification according to one embodiment of the invention. The amplifier or central controller performs an automatic classification 111, i.e. starts determining classes of loads on every single amplifier output or at least on a number of user defined outputs. When the amplifier or computer connected to the amplifier has finished the classification, the result 112 will be presented to the user e.g. on a display mounted on the amplifier or on a computer connected to the amplifier. The primary result of the automatic classification is an indication of which loads from among a predefined set of loads, e.g. from a database, are connected to which amplifier output. In case of an unknown or unclassifiable load on a specific channel, the result will in a simple embodiment merely contain that the load on that channel cannot be determined. Other secondary results can among other things be a determination or estimation of the number of loads coupled in parallel to one channel of the amplifier, a reference impedance curve for use in subsequent live monitoring of the loads, or any other information derivable from the measured characteristics and obtainable from the database by cross-referencing with the determined load type or model, e.g. information about rated power handling, temperature handling, etc.

Fig. HB illustrates a flow diagram of the principle of automatic classification according to an alternative embodiment of the invention. The principle illustrated in fig HB differs from the principle in fig HA by the output step 114 providing more information to the user than the output step 111 of fig. 1 IA. Such extra information may e.g. regard plain information such as the probabilities of the load classifications being correct or detailed information about the loads or it may regard action- demanding information such as options for the user to choose from, or action points to carry out. Examples of options for the user to choose from may be providing the user with 2 or more probable load class matches for each channel or a number of problematic channels and let the user tell the system which class from among the few probable options is correct, etc. Examples of action points for the user to carry out may be providing the user with information about apparently significantly worn loads and have the user do a manual inspection, etc. In any case, the user may be able to accept the result as it is, or input information or change the connections, and have a new classification carried out to reflect any changes. In an advanced embodiment of the invention, the amplifier will not provide a power signal to a load which it does not know, or which seems to be excessively worn or damaged, or which does not match the output channel type and power rating, before the user has actively confirmed to the system that the connection is deliberate and desired. Fig. 12A illustrates a flow diagram of the principle of automatic verification according to one embodiment of the invention. The user starts with a step of selecting loads 121, which may as described above comprise browsing through available loads and selecting one for each connected output channel, or e.g. by uploading a system configuration plan to a central network controller. Also as described above, the user may in an advanced embodiment input reference features or classifiers for otherwise unknown loads in order for the system to be able to verify such loads at the output channels.

Then an automatic verification 122 is carried out where the amplifier starts testing loads on every single amplifier output or at least on a number of user defined outputs, with regards to the degree of correspondence with the expected classes predefined by the user in the first step. When the amplifier or computer connected to the amplifier has finished the verification, the result 123 will be presented to the user e.g. on a display mounted on the amplifier or on a computer connected to the amplifier, e.g. a laptop computer connected to a wireless data network to which also the amplifiers are connected. The primary result of the automatic verification is an indication of the output channels where the actual load belongs to the load class predefined by the user. Other secondary results can among other things be a verification of whether the number of loads coupled in parallel to one channel of the amplifier corresponds to the expected number, or e.g. a reference impedance curve for use in subsequent live monitoring of the loads, or any other information derivable from the measured characteristics and obtainable from the database by cross- referencing with the verified load type or model, e.g. information about rated power handling, temperature handling, etc.

Fig. 12B illustrates a flow diagram of the principle of automatic verification according to an alternative embodiment of the invention. The principle illustrated in fig 12B differs from the principle in fig 12A by the output step 126 providing more information to the user than the output step 123 of fig. 12A. Such extra information may e.g. regard plain information such as the probabilities of the load verifications being correct, or detailed information about the loads or it may regard action- demanding information such as options for the user to choose from, or action points to carry out. Examples of action-demanding information may e.g. be providing the user with information about a problematic verification and have the user do a manual verification, etc. In any case, the user may be able to accept the result as it is, or input information or change the connections, and have a new verification carried out to reflect any changes. In an advanced embodiment of the invention, the amplifier will not provide a power signal to a load which it cannot verify as being the expected load, or which seems to be excessively worn or damaged, or which does not match the output channel type and power rating, before the user has actively confirmed to the system that the connection is deliberate and desired.

As mentioned above, the user decisions 115 or 127 may further comprise inputting data from which a cable component or cable impedance can be determined.

Fig. 13A illustrates an embodiment of the present invention. Several amplifiers 1 with the capability to determine a class of a load according to the present invention exist. They may be part of a single setup, owned by the same company, or they may be owned by different users, located at different places and using them for different purposes with a different data port 130. By means of a data medium 131, e.g. a flash memory stick, suitable for use with the data port 130, classifiers or measured data may be transferred to and from the amplifiers. A central data storage CD is provided, also comprising a data port 130.

Fig. 13B illustrates a preferred embodiment of the present invention. Several amplifiers 1 with the capability to determine a class of a load according to the present invention exist. They may be part of a single setup, owned by the same company, or they may be owned by different users, located at different places and using them for different purposes with different loads. Each amplifier comprises a data port 130. Some amplifiers may be connected in a data network 132, e.g. the Internet, a LAN, a mobile network, etc., e.g. by cabled network connections 133, wireless network connections 134, or any other suitable connection means. Some amplifiers may not be connected but requires a data medium 131 to transfer data. Such data may be provided to the data network 132 by means of a laptop or PC with a suitable data port and a suitable network connection. One or more central data storages CD may also preferably be connected to the data network 132. In an embodiment of the invention, the network further comprises one or more classifier providers CP, which are companies dedicated to establishing classifiers and distributing them to the central data storages CD or amplifiers 1.

The embodiments shown in fig. 13A and 13B can be thought of as communities for distributing classifiers or data related to classifiers, requesting classifiers, or verifying classifiers. The central data storage CD may e.g. be able to receive classifiers or data from the classifier providers CP or from the amplifiers 1. The amplifiers may e.g. be able to receive classifiers from the central data storage CD, the classifier providers CP or directly from other amplifiers.

Claims

Patent claims

1. Method of determining a class of a load by

- providing at least one classifier representing at least one class, - measuring and extracting at least one feature (EFC) of a measured load (LS; 2, 6),

- determining the class (CL) of said measured load statistically on the basis of said at least one feature and said at least one classifier.

2. Method of determining a class of a load according to claim 1 comprising the steps of

- providing at least one classifier representing at least one class,

- measuring and extracting at least one feature (EFC) of a measured load (LS; 2, 6) connected to an amplifier (1),

3. Method of determining a class of a load according to claim 1 or 2, wherein the method is applied within an amplifier (1) and wherein said measured load (LS; 2, 6) is connected electrically to said amplifier (1).

4. Method of determining a class of a load according to any of the claims 1 to 3, wherein the amplifier executes the method by means of data processing means (3) according to instructions stored in memory means.

5. Method of determining a class of a load according to any of the claims 1 to 4, wherein the amplifier comprises a data port by means of which models may be transferred to or from the amplifier.

6. Method of determining a class of a load according to any of the claims 1 to 5, wherein said classifier or data related with said classifier is stored in a data storage

(5).

7. Method of determining a class of a load according to any of the claims 1 to 6, wherein said amplifier (1) comprises said data storage (5).

8. Method of determining a class of a load according to any of the claims 1 to 7, wherein said data storage (5) is comprised by a central data storage to which said amplifier (1) has irregular or continuous access.

9. Method of determining a class of a load according to any of the claims 1 to 8, wherein said method comprises a decision layer.

10. Method of determining a class of a load according to any of the claims 1 to 9, wherein said at least one classifier comprises a parametric classifier.

11. Method of determining a class of a load according to any of the claims 1 to 10, wherein said at least one classifier comprises a non-parametric classifier.

12. Method of determining a class of a load according to any of the claims 1 to 11, wherein said at least one class is established by providing a set of reference features (RFC) of a plurality of reference loudspeaker units (RLS).

13. Method of determining a class of a load according to any of the claims 1 to 12, wherein at least one of said reference features (RFC) is provided as a function of frequency.

14. Method of determining a class of a load according to any of the claims 1 to 13, wherein at least one of said features (EFC) is determined as a function of frequency.

15. Method of determining a class of a load according to any of the claims 1 to 14, wherein said determination of a class of said load (LS; 2,6) is based on a probability of match between said measured and extracted features (EFC) of said measured load and one or several reference features (RFC).

16. Method of determining a class of a load according to any of the claims 1 to 15, wherein said determination of a class of said load (LS; 2,6) involves a statistical determination.

17. Method of determining a class of a load according to any of the claims 1 to 16, wherein said determination of a class of said load (LS; 2,6) involves a non-linear signal processing.

18. Method of determining a class of a load according to any of the claims 1 to 17, wherein the determination is performed automatically.

19. Method of determining a class of a load according to any of the claims 1 to 18, wherein the determination is performed semi-automatically.

20. Method of determining a class of a load according to any of the claims 1 to 19, wherein the determination is performed during a calibration phase.

21. Method of determining a class of a load according to any of the claims 1 to 20, wherein the determination is performed during a verification phase.

22. Method of determining a class of a load according to any of the claims 1 to 21, wherein said determination of a class of said load (LS; 2,6) involves indication of one or more of the make, model, band, driver and/or number of parallel coupled loudspeakers of the measured load.

23. Method of determining a class of a load according to any of the claims 1 to 22, wherein said determination of a class of said load (LS; 2,6) involves indication of the type of the measured load.

24. Method of determining a class of a load according to any of the claims 1 to 23, wherein said determination of a class of said load (LS; 2,6) involves indication of possible classes to a user and where the possible classes are associated with indication of calculated probabilities.

25. Method of determining a class of a load according to any of the claims 1 to 24, wherein said determination of a class of said load (LS; 2,6) involves indication of the load being a predefined load, or the probability of the load being a predefined load.

26. Method of determining a class of a load according to any of the claims 1 to 25, wherein said features (EFC) are extracted in at least two separate or overlapping feature extraction bands (B 1 , B2, ... , B7).

27. Method of determining a class of a load according to any of the claims 1 to 26, wherein said features (EFC) are extracted in at least three overlapping feature extraction bands (B 1 , B2, ... , B7).

28. Method of determining a class of a load according to any of the claims 1 to 27, wherein said features (EFC) are extracted in at least seven overlapping feature extraction bands (B 1 , B2, ... , B7).

29. Method of determining a class of a load according to any of the claims 1 to 28, wherein the determination of a class of said measured load is applied for deriving of information associated with said class (CL).

30. Method of determining a class of a load according to any of the claims 1 to 29, wherein said at least one feature (EFC) comprises at least one electrical measurement or is derived from at least one electrical measurement.

31. Method of determining a class of a load according to any of the claims 1 to 30, wherein said at least one feature (EFC) comprises impedance of the load as function of frequency, variance and/or mean value of the impedance of the load, resonance frequency of the load, DC resistance of the load, etc.

32. Method of determining a class of a load according to any of the claims 1 to 31, wherein said measured and extracted features are compensated for a cable component before said determination of class is performed on the basis of said features.

33. Method of determining a class of a load according to any of the claims 1 to 32, wherein at least two features (EFC) are established and analyzed in different feature extraction bands (Bl, B2, ..., B7), overlapping or non-overlapping.

34. Load class determining amplifier, comprising an amplifier (1), a data processing means (3) and an amplifier output (AO; 4) connected to a load (LS; 2, 6), said amplifier (1) comprising means for measuring and extracting at least one feature (EFC) of said load, and said data processing means (3) comprising means for determining the class (CL) of said load statistically on the basis of said at least one feature and at least one classifier.

35. Load class determining amplifier according to claim 34, comprising means for carrying out a method of determining a class of a load according to any of the claims 1 - 33.

36. Method of verifying if a load (LS; 2, 6) connected to an amplifier output (AO; 4) corresponds to a predefined load, said method comprising the steps of

- measuring and extracting at least one feature (EFC) of a measured load (LS; 2, 6),

- determining statistically on the basis of said at least one feature (EFC) and at least one classifier representing at least one class if said measured load belongs to the class of said predefined load.

37. Method of verifying if a load corresponds to a predefined load according to claim 36, comprising any of the features of claims 1 to 33.

38. Load verification amplifier comprising data processing means (3) for carrying out a method of verifying if a load corresponds to a predefined load according to any of the claims 36 - 37.

39. System comprising an amplifier (1) according to claim 34, 35 or 38 and at least one load (2, 6).

40. Use of a method according to any of the claims 1 - 33 or 36 - 37.