EP0712263B1 - Programmable hearing aid - Google Patents

Description

The invention relates to a programmable hearing aid with an amplifier and transmission part which can be adjusted in terms of its transmission properties between the microphone and the listener to different transmission characteristics.

From WO 91-08654 a programmable hearing aid is known with an amplifier and transmission part which can be adjusted in terms of its transmission properties between the microphone and the listener to different transmission characteristics. Several different transmission functions are stored in a memory of a control unit and can be selected by a user to adapt the hearing aid to an acoustic environment.

A simulator is known from EP 0 540 168 A2 in which the properties of a device to be controlled, e.g. for water supply, are modeled. The simulator comprises a neural network which can be learned in relation to a real control parameter of the simulated device. This improves the quality of the simulation of the real device.

From US 5,101,361 an analog hardware of a learnable neural network is known, in which a number of programmable resistance memory elements serve as synaptic connection elements, which form an adjustable, weighted path between the neurons. The patent discloses an implementation of the delta back propagation algorithm in analog VLSI technology.

A solar-powered battery charger for rechargeable hearing aid batteries is known from US Pat. No. 5,253,300. Contacts on the charger and on the housing of the hearing aid allow recharging without removing the batteries from the hearing aid housing possible. One embodiment of the charger comprises a device that monitors the state of charge of the battery.

In ELEKTRONIK, Vol. 41, No. 2, January 21, 1992 GERMANY, pages 100-101, G. Trautzl "NEURONAL NETWORK SUPPORT FUZZY LOGIK TOOL" describes a data processing system by means of which neural networks automatically fuzzy logic rules can be generated. There is thus a link between various artificial intelligences.

In a hearing device known from EP-A-0 064 042, eight parameter sets for different transmission characteristics for different environmental situations are stored in a hearing device memory. The various parameter sets for the eight stored programs can be called up one after the other by pressing a switch. A control unit controls a signal processor connected between the microphone and the handset, which then sets a first transmission function intended for an intended environmental situation. However, the stored signal transmission programs can only be called up one after the other via the switch until, in the opinion of the hearing device wearer, the transmission function that is just right for the given environmental situation has been found. On the other hand, according to EP-B-0 064 042, an automatic switchover to another transfer function should also be provided if the user comes from a noisy environment to a quiet environment, for example, or vice versa. Without a solution being specified for the automatic switchover, this switchover should take place cyclically like the manual switchover. If you want to set other than the stored transfer functions, the non-volatile memory must be erased by an external programming unit and re-programmed by this. In this way, it is also possible to adapt the programmable hearing aid to changing hearing damage.

Accordingly, in the known programmable hearing aid, a plurality of parameter sets, so-called hearing situations, which can be selected by the user, are generally stored. Each of these parameter sets represents the sensibly coordinated setting of all signal processing parameters for a specific acoustic situation, e.g. at rest, i.e. without disturbing background noise or a conversation situation with low-frequency noise, etc. The hearing aid wearer selects the desired situation by pressing a button on the hearing aid.

A largely automatic selection of the hearing situation required in each case or the automatic setting of individual signal processing parameters by evaluating the signal according to rules to be specified is now desirable.

The object of the invention is to provide a hearing aid which is characterized by a control system which enables the desired automatic or largely automatic adaptation dependent on input or measurement signals.

According to the invention, this object is achieved in a programmable hearing device of the type mentioned at the outset by a neural structure which is associated with the signal path from the microphone to the listener and to which a data carrier is assigned, signals being tapped from the signal path from one or more tapping points and fed to a module for signal processing and the processed signals can be fed to the neural structure, which generates control signals for the automatic signal-dependent selection of hearing situations stored in the data memory.

In an advantageous embodiment of the invention, the neural structure forms a controller for the automatic signal-dependent selection of hearing situations stored in the data memory. An essential feature of the hearing device according to the invention is therefore a controller based on the principle of the neural structures for the automatic signal-dependent selection of hearing situations or for setting individual signal processing parameters in the hearing device. This basic principle allows any selection behavior to be set. Analog circuit technology lends itself to the implementation of the principle of neural structures in terms of circuitry.

Advantageous embodiments of the invention are characterized by the claims.

Further advantages and details of the invention are explained in more detail below with reference to the exemplary embodiments shown in the figures.

• Figur 1 ein Blockschaltbild eines erfindungsgemäßen Hörgerätes,
• Figur 2 ein Blockschaltbild eines Controllers nach dem Prinzip einer neuronalen Struktur eines Hörgerätes gemäß Figur 1,
• Figur 3 ein Modul zur Signalaufbereitung nach Figur 1,
• Figur 4 ein Blockschaltbild eines einzelnen Neurons,
• Figuren 5a, 5b, 5c Beispiele für mögliche Schwellenwertverläufe der Ausgabefunktion W gemäß Figur 4,
• Figur 6 ein einlagiges, rückgekoppeltes Netz mit beispielhafter Verschaltung von drei Neuronen,
• Figur 7 ein mehrlagiges, rückkopplungsfreies Netz mit beispielhafter Verschaltung von elf Neuronen in drei Lagen,
• Figur 8 ein Schaltungsbeispiel für die schaltungstechnische Realisierung eines einlagigen rückgekoppelten Netzes gemäß Figur 6,
• Figur 9 eine mögliche Schaltung zur Realisierung einer Synapse mit programmierbarer Verbindungsstärke,
• Figur 10 eine Ausführung einer Schaltung für eine Synapse mit programmierbarer variabler Verbindungsstärke,
• Figur 11 ein Ausführungsbeispiel einer dem nicht gezeichneten Signalpfad zwischen Mikrofon und Hörer eines Hörgerätes nebengeordneten neuronalen Struktur, welche mit einem Fuzzy-Logik-System kombiniert ist.

Show it:

• FIG. 1 shows a block diagram of a hearing aid according to the invention,
• FIG. 2 shows a block diagram of a controller based on the principle of a neural structure of a hearing aid according to FIG. 1,
• FIG. 3 shows a module for signal processing according to FIG. 1,
• FIG. 4 shows a block diagram of an individual neuron,
• FIGS. 5a, 5b, 5c examples of possible threshold value profiles of the output function W according to FIG. 4,
• FIG. 6 shows a single-layer, feedback network with an exemplary connection of three neurons,
• FIG. 7 shows a multi-layer, feedback-free network with exemplary connection of eleven neurons in three layers,
• 8 shows a circuit example for the implementation of a single-layer feedback network according to FIG. 6,
• FIG. 9 shows a possible circuit for realizing a synapse with programmable connection strength,
• FIG. 10 shows an embodiment of a circuit for a synapse with programmable variable connection strength,
• FIG. 11 shows an exemplary embodiment of a neural structure which is adjacent to the signal path (not shown) between the microphone and the receiver of a hearing device and which is combined with a fuzzy logic system.

The hearing aid 1 according to the invention, shown schematically in FIG. 1, records 2 sound signals via a microphone. This acoustic information is converted into electrical signals in the microphone. After signal processing in an amplification and transmission part 4, the electrical signal is fed to a receiver 3 as an output converter.

In order to eliminate the need for an additional microphone or another sensor, signals 8 are tapped from the signal path of the hearing device 1 between its microphone 2 and its receiver 3 at certain, desired tapping points. These signals 8 are fed to a neural structure 5 arranged next to the signal path in the hearing aid. The signals 8 first arrive at a module 9 for signal processing and from its outputs processed signals 10, 10 ', 10 '' of the neural structure 5. A data carrier 6 is assigned to the neural structure 5, in which configuration information of the neural structure is stored. Taking into account the configuration information of the data carrier 6, the neural structure 5 generates control signals 11 from the processed signals 10, 10 ′, 10 ″, which can be fed to the amplifier and transmission part 4 in order to adapt its transmission characteristics. These control signals 11 can be used to select parameters of the amplifier and transmission part stored in a data memory 12 assigned to the signal path of the hearing aid or to change the amplifier and transmission characteristics to the amplifier and transmission part 4.

As also shown in FIG. 2 for the structure of a controller, 7 signals are tapped from the signal path of the hearing aid at all relevant points or tapping points and processed in a suitable manner. These processed signals and any other system information, e.g. whether microphone or telephone operation is desired are added to the "Neural Structure" block. The outputs of the controller are signals which effect the setting of the listening situations or which represent the setting variables of individual signal processing parameters. The behavior of the controller does not necessarily have to be unchangeable (i.e. completely described by the hardware structure), but can be configurable (e.g. by programming). Configuration information of the controller can be stored in a memory in the hearing aid.

According to a first embodiment, the neural structure 5, together with the module 9 for processing the signals and the data carrier 6, then forms a controller for the automatic signal-dependent selection of hearing situations stored in the data memory 12.

According to a second embodiment it is provided that the neural structure 5 forms a controller for the automatic setting of individual signal processing parameters of the amplifier and transmission part 4.

In a further embodiment of the hearing aid according to the invention, means 13 are provided for detecting system states of the hearing aid 1, the output signals 14 of which can be fed to the module 9 for signal processing and / or the neural structure 5, these output signals being able to be taken into account when generating the control signals 11.

Control elements which can be actuated by the hearing aid wearer, such as switches, buttons, potentiometers or the like, can be provided as means 13 for detecting system states of the hearing aid 1. Then the hearing aid can e.g. be equipped with a situation switch which enables the hearing aid wearer - as described at the beginning of EP-B-0 064 042 - to choose a stored transmission function which he thinks is suitable for the given environmental situation. On the other hand, the hearing aid can e.g. have a switch for switching from microphone operation to telephone coil operation. Furthermore, the hearing aid regularly has a volume control with which the hearing impaired influences the volume.

Finally, the hearing device can also be characterized in that a device 15 which monitors the state of charge of the hearing device battery, not shown, is additionally provided as a means for detecting system states of the hearing device 1. Thereafter, it is possible that the respective state of charge of the hearing aid battery is also taken into account when generating the output signals 11 of the neural structure 5 or the controller, which can also be determined for individual hearing situations and / or signal processing parameters.

In the known so-called multi-channel devices, i.e. in hearing aids with several frequency channels, according to the invention the signals 8 are tapped from the signal paths formed by the individual channels. As shown in Figure 3, the signals 8 from the signal paths and the output signals 14 of the means 13, 13 'for detecting system states of the hearing aid e.g. by rectification, the formation of suitable temporal averages, and possibly prepared from their derivatives. In signal processing - module 9 - at least one input variable, e.g. Signals 8, 14, signals 10 processed by a component 16 for rectification and / or a component 17 for averaging (forming a time average) and / or a component 18 for time derivation (derivation block d / dt) from the signals 8, 14, 10 ', 10' 'which are fed to the neural structure 5. Likewise, the signals 14 can also be fed directly to the neural structure (FIG. 2).

Examples for realizing the neural structure are described with reference to FIGS. 4-10.

Neural structures consist of many similar elements or neurons 19. The function of the neural structure as a whole essentially depends on the way in which these neurons are interconnected.

• Propagierungsfunktion U:u (t) =Σe_i (t) *w_i Die Ausgangsgröße dieser Funktion ist die Summe aller, jeweils mit dem individuellen Faktor w_i multiplizierten Eingangssignale.
• Aktivierungsfunktion V:v(t)=f(u(t)) Im allgemeinen Fall geht in die Ausgangsgröße auch deren eigene Vorgeschichte ein. In vielen Fällen kann hierauf jedoch verzichtet werden. v(t) zum Zeitpunkt t=t₀ ist dann nur noch eine Funktion von u(t) zum Zeitpunkt t=t₀.
• Ausgangsfunktion W:w(t) Sie nimmt eine Schwellenwertbildung vor. Dabei sind gemäß Figur 5 zwei grundsätzliche Arten der Schwellenwertbildung möglich.

FIG. 4 shows the block diagram of an individual neuron 19. The neuron generates the output signal a _j (t + ΔT) at the time t + ΔT from theoretically any number of input signals e _i (t) at the time t. Its function can be broken down into three basic functions:

• Propagation function U: u (t) = Σe _i (t) * w _i The output variable of this function is the sum of all input signals multiplied by the individual factor w _i .
• Activation function V: v (t) = f (u (t)) In the general case, the output quantity also includes its own history. In many cases, however, this can be dispensed with. v (t) at time t = t ₀ is then only a function of u (t) at time t = t ₀ .
• Output function W: w (t) It creates a threshold value. According to FIG. 5, two basic types of threshold value formation are possible.

According to FIG. 5a, the course of the output function W represents a step function at the threshold value s.

According to FIGS. 5b and 5c, the output function W has a continuous course around the threshold value s. FIG. 5b shows a continuous, so-called sigmoid curve of the output variable with limitation to a maximum and a minimum output value. A frequently used characteristic is the sigmoid: w (t) = 1 / (1 + exp (- (v (t) -s))). FIG. 5c shows a linear course in the transition area.

The signals which are processed by the neural structure can be designed as voltage signals, current signals or as frequency-variable pulse signals. In the latter case, the signal may have to be converted into a continuous current or voltage signal and back again at some points in the neural structure with the aid of suitable circuits.

FIG. 6 shows the exemplary connection of three neurons 19 to the typical structure of a single-layer feedback network with the inputs e _i (t) and the outputs a _j (t + ΔT).

Figure 7 shows an example of the structure of a multilayer feedback-free network. Depending on the one to be implemented The function of the neural structure is to use one or the other network structure. Mixed forms of both structures are also possible.

The function of a neural structure as a whole is essentially determined by the network structure and by the weighting functions of the input signals on each neuron 19. These parameters can be permanently set through the implementation in terms of circuitry if constant behavior is desired. If, on the other hand, a change in behavior should be possible, some or all of these parameters must be programmable. Their respective values must then be stored in a configuration memory or data carrier 6. The individual memory elements can be arranged in a concentrated form or locally assigned to the respective neuron.

The stored parameters can be modified either by external programming of the memory elements and / or by an algorithm implemented in the circuit. Modification is also possible here while the neural structure is in operation.

Figure 8 shows an example of the circuitry implementation of a single-layer feedback network. Amplifiers 24 with complementary outputs act as threshold elements. The connections (synapses) between the outputs and inputs of the neurons are weighted using the guide values R _ij . The addition of the input signals for each neuron (currents I _ij = U _i / R _ij ) takes place in the circuit nodes at the input of each amplifier. The output signals of the amplifiers and thus the neural structure are the voltage signals U _i . E1 to e4 denote the inputs of the circuit and a1 to a4 denote inverting and non-inverting outputs of the circuit.

FIG. 9 shows a possible circuit implementation of a synapse (weighted input of a neuron) with a programmable connection strength. Here only the connection strengths +1, -1 and 0 are possible and the signals to be transmitted from this synapse can only assume the logical values 0 and 1. If both memory cells 25, 26 are programmed so that they block the respective associated switching transistor 27 or 28, output a is independent of input e; the synapse therefore represents an interruption (connection strength 0). If, on the other hand, the memory cell 25 is programmed to close the switch and the memory cell 26 to open the associated switch, then a current flows from output a (logic 1) when the input is logic 1, and no current (logic 0) if the input is logic 0. So the synapse acts as a connection of strength +1. If both memory cells 25, 26 are programmed inversely for this purpose, the inverse logic behavior results. The synapse then acts as a connection of strength -1. V _dd indicates the circuit connection to the supply voltage in the drawing.

FIG. 10 shows a possible implementation of a programmable synapse with variable connection strength. It works on the principle of the multiplier. The strength of the synaptic connection is stored as the difference between two analog voltage values on two capacitors 29, 30. The output signal (current I _out ) is the product of the input signal (voltage V _in ) multiplied by the voltage difference stored on the capacitors (V _W = V _{W +} -V _W- ). If the voltages V _{W +} and V _W- are stored on the floating gates of corresponding EEPROM transistors, permanent storage of the synapse strength is also possible.

An advantageous further development of the invention is characterized in that the controller 5, which is arranged next to the signal path from the microphone 2 to the receiver 3, has additional functional parts comprises, which operate according to a fuzzy logic system 20. According to FIG. 11, the fuzzy logic system 20 has components for fuzzification 21 and for inference formation 22 and, furthermore, an assigned decision means component 23 is designed as a neural structure. According to the exemplary embodiment shown in the block diagram, in addition to its signal path (not shown here) from the microphone via the amplifier and transmission part to the listener, the hearing aid has a controller 5 as a neural structure with a data carrier 6 for configuration information of the controller and a signal conditioning unit 9. Signals 8 from signal path of the hearing aid are fed to signal processing unit 9. Signals 14 emanating from the means 13, 13 'about the system state of the hearing aid can be fed to the signal processor 9 and / or the controller 5. The controller 5 comprises a fuzzy logic system 20 which is subdivided into a component for fuzzification 21 and a component for inference formation 22. The processed signals 10, 10 ', 10''of the signal processing 9 and optionally signals 14 of the fuzzification component 21 are supplied. The inference formation component 22 is connected downstream of the fuzzification component 21. The further decision means component 23 of the controller 5 is then designed as a neural structure. All three components 21, 22, 23 are exchanging information with the data carrier 6. The generated control signals 11 form, for example, selection signals for individual listening situations and / or signal processing parameters.

A further advantageous realization of the controller with a combination of the principles of the neural structures and the fuzzy logic can consist in that the implementation of the rules, which determine the regulation or selection behavior of the fuzzy logic part, is realized with the aid of a neural structure ,

The principle of operation as well as a possible circuit implementation of the blocks fuzzification and inference formation is described in the European patent application 94104619.5. The neural structure for 1-out-of-N selection corresponds in principle to the structure described in the earlier application, but is more advantageously solved.

Significant advantages of the invention result in a better adaptation of the hearing aid to the hearing damage through continuous signal-dependent setting of signal processing parameters and by relieving the burden on the hearing aid wearer due to the automatic selection of signal processing parameters or sets of signal processing parameters (hearing situations).

Claims (14)

1. Programmable hearing aid (1) with an amplification and transmission component (4), the transmission properties of which can be adjusted between microphone (2) and receiver (3) to different transmission characteristics, characterised by a neural structure (5) parallel to the signal path from the microphone to the receiver, a data medium (6) being assigned to the said structure, with signals (8) being tapped from the signal path from one or more tapping points (7) and supplied to a module (9) for signal processing, and it being possible to supply the processed signals (10, 10', 10") to the neural structure (5), which generates control signals (11) for the automatic signal-dependent selection of hearing situations stored in a data memory (12) assigned to the signal path.
2. Hearing aid according to Claim 1, characterised in that the neural structure (5) forms, together with the module (9) for processing the signals and the data medium (6), a controller for automatic adjustment of individual signal processing parameters of the amplifier and transmission component (4).
3. Hearing aid according to Claim 1 or 2, characterised in that means (13) for recording system states of the hearing aid (1) are provided, the output signals (14) of which can be supplied to the module (9) for signal processing and/or to the neural structure (5), it being possible to take these output signals into account when generating the control signals (11).
4. Hearing aid according to Claim 3, characterised in that control elements, for example switches, pushbuttons and potentiometers, which can be actuated by the hearing aid wearer are provided as means (13) of recording system states of the hearing aid (1).
5. Hearing aid according to Claim 3, characterised in that a device (15) which monitors the discharge degree of the hearing aid battery is provided as means (13, 13') for recording system states of the hearing aid (1).
6. Hearing aid according to one of Claims 1 to 5, characterised in that the module (9) for signal processing contains components for rectification (16) and/or for averaging (17) and/or for time derivation (18).
7. Hearing aid according to one of Claims 1 to 6, characterised in that in the case of a hearing aid with a plurality of frequency channels tapping points (7) are provided for the signals (8) in the individual channels.
8. Hearing aid according to Claim 1, characterised in that configuration information of the controller is stored in the data medium (6) which is assigned to the neural structure (5).
9. Hearing aid according to Claim 1, characterised in that the controller parallel to the signal path from the microphone to the receiver includes additional functional devices (20) which work on the principle of fuzzy logic.
10. Hearing aid according to Claim 9, characterised in that the fuzzy logic system (20) has components for fuzzification (21) and for inference formation (22), an assigned decision-making component (23) being formed as a neural structure.
11. Hearing aid according to one of Claims 1 to 10, characterised in that the neural structure (5) is implemented either as a single-layer feedback network (Figure 6) or as a multi-layer feedback-free network (Figure 7) or as a hybrid form of both network structures.
12. Hearing aid according to one of Claims 1 to 11, characterised in that the weighting functions at the input of all neurons are permanently predetermined by the circuit structure.
13. Hearing aid according to one of Claims 1 to 11, characterised in that the weighting functions at the input of all neurons are executed on a programmable basis by an external controller, the programming data being stored in a shared data medium (6) or the respective programming data being stored in individual memory modules assigned to the neurons.
14. Hearing aid according to one of Claims 1 to 11, characterised in that the weighting function at the input of all neurons can be modified by an algorithm implemented in the circuit structure at particular moments in time or continually.