T 0837/12 (Combining audio compression and feedback cancellation in a hearing aid/GN RESOUND) 30-03-2016

1. General

The present patent concerns a hearing aid comprising audio processing means, including audio compression means, in a forward path between a microphone and a speaker, and feedback cancellation means in a reverse path, which aim to estimate and cancel the physical audio feedback signal by subtracting an estimated signal from the signal output by the microphone. Very broadly stated, the patent discloses a concept concerned with, inter alia, providing information from the feedback cancellation means to the compression means in order to prevent loop instability, and thus prevent whistling or howling of the hearing aid.

2. Main request - claim 1 - novelty with respect to D9

2.1 It was not disputed that D9 discloses all the features of claim 1 except for the features:

"said hearing aid processing means includes compression means (40) for performing audio compression; said hearing aid further comprising:

means (406) for providing information from the feedback cancellation means to the compression means, and wherein said compression means adjust its [sic] operation based upon information provided by the feedback cancellation means".

2.2 The respondent argued that these features were also disclosed in D9. In this respect, the respondent argued that the multiplication circuit 211 combined with the limiter circuit 15 in Fig. 2 performed the same function as a compression circuit, which was to make soft sounds louder without making loud sounds louder.

2.3 However, in the board's view, neither the multiplication circuit 211 nor the limiter 15, nor the combination, performs the function of a compression circuit.

As regards the multiplication circuit 211, this is controlled to reduce the gain when the loop gain exceeds a predetermined value K (cf. page 10, lines 19 to 26). This control of the multiplication value is not based on the level of the input signal, i.e. is not controlled in order to "make soft sounds louder without making loud sounds louder", and therefore not related to audio compression.

2.5 As regards the limiter circuit 15, the board notes that limiter circuits are of two broad types:

(a) circuits which merely clip the signal to a desired maximum level; and

(b) circuits where the gain is controlled in order to prevent the signal exceeding a desired level.

A type (b) limiter functions in a manner similar to a compression circuit, as explained by the respondent in the letter dated 23 December 2013, cf. page 3, last paragraph ("Typically there will be a level detector and a mechanism for adjusting the gain depending on the level detected."). Therefore, a type (b) limiter arguably is a compression circuit. However, a type (a) limiter has no gain control mechanism and is clearly not a compression circuit.

2.6 As regards D9, the limiter is apparently of type (a), i.e. not a compression circuit. This follows from the reference in D9 to the limiter disclosed in document D2 (cf. D9, page 5), which is clearly of type (a), i.e. not a compression circuit (cf. D2, Fig. 5 and col. 6, lines 38-48); see also the statement of grounds of appeal, middle of page 3).

2.7 As neither the gain circuit nor the limiter provides a compression function, neither does the combination of these elements.

2.8 As D9 does not disclose a compression circuit, the subject-matter of claim 1 is new with respect to the disclosure of D9 (Articles 52(1) and 54 EPC).

3. Main request - claim 1 - inventive step with respect to the combination of D1 and D9

3.1 It was not in dispute that D1, in particular Fig. 15 and the associated text, discloses all the features of claim 1 with the exception of the features:

"means for providing information from the feedback cancellation means to the compression means, and wherein said compression means adjust its [sic] operation based upon information provided by the feedback cancellation means".

3.2 The problem to be solved by these features is to prevent the loop consisting of the [forward] processing means and the feedback cancellation means becoming unstable.

D9 provides a solution to this problem (cf. page 11, lines 10 to 17) consisting of determining the loop gain and reducing the forward gain if the loop gain reaches a threshold value (cf. page 11, lines 17 to 26). The loop gain is obtained from the filter coefficients of feedback cancellation means FIR 27 (cf. page 11, lines 26 to 32 and page 12, lines 7 to 28). In the board's view, the skilled person would attempt to apply this same solution to the system of D1, Fig. 15 to counter the problem of loop instability. An obvious way to implement this solution that would occur to the skilled person is to control the gain of the "hearing aid signal processing" block (i.e. the compression circuit) in Fig. 15 of D1 based on the filter coefficients from the "Filter estimator". The skilled person would thus arrive at the subject-matter of claim 1 without inventive skill.

3.4 The appellant presented the following counter-arguments at the oral proceedings:

(i) The skilled person would not look to D9 as it provides for feedback cancellation using a noise generator which injects wide band noise in order to correctly adapt the FIR filter 27. The only way to apply the teaching of D9 would be to include the noise generator, since without it, it would be impossible to calculate a good estimate of the open loop gain due to the limited spectrum of the feedback signal. However, the skilled person would reject any solution using a noise generator, since such hearing aids were known to be uncomfortable for the wearer. This would thus be a retrograde step that the skilled person would not envisage.

(ii) D1 discloses a system with stability problems since the forward gain path is non-linear. The skilled person would not look to D9 for a solution as it describes a system with linear gain in the forward path.

Re (i): The board is unconvinced that the skilled person would reject D9 on the basis that it includes a noise generator, and moreover does not agree that any solution based on D9 must include a noise generator. In this respect, the above-mentioned section of D9 (i.e. page 11, lines 17 to 26) describing a solution to the aforementioned technical problem is essentially self-contained and makes no reference to the noise generator. Furthermore, in accordance with D9, it is only necessary to carry out an approximate calculation of the actual loop amplification (cf. page 11, lines 17 to 21), i.e. accuracy is not essential. Self-evidently, a lack of accuracy could easily be taken into account by choosing a conservative value of K, the constant which regulates when to reduce the hearing aid's amplification (cf. page 10, lines 19 to 23, and page 12, lines 7 to 28).

Re (ii): The board sees no inherent issue with non-linearity that would dissuade the skilled person from applying the approach of D9. A greater tendency to instability can again be taken into account when choosing the value K.

3.6 In the written proceedings, the appellant argued that if D9 were combined with D1, a gain element would be placed at the output of the compression circuit. Consequently, there would be no control of the compression circuit itself.

However, the board notes that the compression circuit has gain calculation elements in each frequency band (cf. Fig. 8). Consequently, it would occur to the skilled person that it would be more efficient to reduce the gain of these elements rather than to introduce an extra gain circuit. However, even in the case that a further gain element were placed after the compression circuit, this gain element could still be regarded as part of the "compression means" defined in claim 1.

The board therefore finds the appellant's arguments unconvincing.

3.7 The board concludes that the subject-matter of claim 1 does not involve an inventive step (Articles 52(1) and 56 EPC).

4. First auxiliary request - claim 1 - amendments (Articles 123(2) and 123(3) EPC)

4.1 Claim 1 of the first auxiliary request differs from claim 1 of the main request in that it additionally includes the features:

"wherein the feedback cancellation means includes a zero filter (212); the hearing aid includes means for calculating a norm of a vector of coefficients of the feedback cancellation means zero filter, and the compression means modifies a gain value based on the norm".

4.2 These added features result in a further limitation of claim 1 as granted, and hence the claim complies with Article 123(3) EPC.

4.3 This added wording is based verbatim on claim 7 as originally filed. Consequently, compliance with Article 123(2) EPC is not in doubt and was not contested.

5. First auxiliary request - claim 1 - interpretation

5.1 The wording added to claim 1 was already included in claim 6 as granted. Consequently, the board is not empowered to examine this amendment for compliance with Article 84 EPC (cf. G 3/14). Nevertheless, certain terms require interpretation before the claim can be examined for inventive step:

A "zero filter" is understood as a filter comprising only zeroes, i.e. no poles, also known as an "all-zero filter".

A "norm of a vector of coefficients", in accordance with the normal meaning of the term "norm of a vector", is regarded as a value representing the length of the vector of coefficients calculated from the square root of the sum of the squares of the coefficient magnitudes. The respondent argued that the term "norm" was vague, and embraced the magnitude of a single coefficient of the vector. This had also been the view of the opposition division (cf. the impugned decision, page 10, third paragraph). The board however notes that "coefficients" in the term "norm of a vector of coefficients" is a plural entity. Consequently, there are at least two coefficients involved in the calculation of the norm.

6. First auxiliary request - claim 1 - inventive step

6.1 It was not disputed that D9 discloses a "zero filter", since the FIR filter 27 in Fig. 2 is implicitly an all-zero filter. It was also not in doubt that the adjustment of the multiplication circuit 211 was based on the coefficients of the FIR filter "FIRCOEF" (cf. page 12, lines 7 to 28).

6.2 D9 is not explicit as to how the filter coefficients are used in the calculation of loop gain to control the multiplication circuit 211. However, D9 clearly teaches that the loop gain is calculated at different frequencies (cf. page 3, lines 20 to 26; page 10, line 35 to page 11, line 3; page 11, lines 17 to 22; page 12, lines 7 to 28).

6.3 The appellant argued at the oral proceedings (see also the statement of grounds of appeal on page 9) that in order to calculate the gain at a given frequency omega, the following expression representing the transfer function of the zero filter has to be evaluated (f0 ... fN-1 being the filter coefficients):

H(jomega) = f0 + f1e**(-jomega) + f2e**(-j2omega) ... + fN-1e**(-j(N-1)omega)

This required a number of calculations at each frequency, which was time consuming and costed power. The coefficient vector norm here would be of no use as it had no frequency component. On the other hand, calculating the norm as claimed was much simpler and led to a more rapid response.

6.4 The respondent argued that in accordance with D9 the skilled person would have to calculate a scalar value from the vector of coefficients. The most natural choice would be the norm of that vector, in particular because this vague term embraced even taking a single coefficient as the norm.

6.5 Considering that D9 requires an open loop gain value to be calculated at specific frequencies, the board is persuaded by the argument of the appellant that D9 teaches away from using a norm of the vector coefficients to control gain, since the norm value calculated from at least two filter coefficients would not be a scalar value related to the gain at a specific frequency. The board can thus see no motivation in D9 to calculate the norm of the vector of filter coefficients. This step is therefore not obvious.

6.6 The board concludes that the subject-matter of claim 1 involves an inventive step with respect to the combination of D1 and D9 (Articles 52(1) and 56 EPC).

7. First auxiliary request - claim 1 - inventive step starting out from D12

7.1 D12 describes a hearing aid with the same basic structure as the hearing aid according to the present patent, in that Fig. 1 shows both compression means ("Feedforward Filter") in a forward path and feedback cancellation means ("Predictor") in a reverse path.

7.2 D12 describes a method for preventing instability based on determining an estimate of the open loop gain K0[k], which is calculated based on the error signal E(k) derived from a DFT of the output of the subtractor which subtracts the cancellation signal from the input signal. The formula for K0[k] is based on formula (13) on page 4. This formula includes no terms based on the coefficients of the predictor.

7.3 However, formula (13) is said to be motivated by formula (11) on page 2 (cf. page 3, right-hand col., last paragraph). This formula includes a term |H˜i[k]|**(2). It follows from col. 2, line 17 that H˜[k] = H - ^H[k], H being the acoustic feedback path and ^H[k] being the predictor transfer function.

The respondent argued that this showed that K0 could be calculated from equation (11) instead of equation (13). Further, in equation (11), H could be replaced by H-^H. The terms of the equation can then be re-formulated such that the modulus of H-^H has to be calculated. By expanding this expression, one arrives at an expression including the term SUM^H**(2), which the respondent considered to be the norm of ^H.

In reply, the appellant argued that equation (11) could not be used in practice because the acoustic feedback path H was unknown.

To this, the respondent argued that although H was not generally known, it could be determined under laboratory conditions such as described in section V of D12, in particular in order to check the reliability of determining K0 using equation (13).

The board however finds the respondent's arguments unconvincing. Firstly, it is pure speculation whether the skilled person would attempt to check the reliability of equation (13) by computing equation (11) under laboratory conditions given a known audio feedback transfer function. Indeed, it is not known whether it would be even possible to know this transfer function accurately. Secondly, it is uncertain whether the test arrangement described in section V can fairly be considered to be a "hearing aid" within the meaning of claim 1, since the system appears to contain processing elements on a DSP board external to a "dummy" hearing aid worn by a "KEMAR-head". Finally, the respondent's mathematical analysis is unconvincing, since even if the skilled person performed the mathematical steps alleged by the respondent, it appears that, rather by chance than design, at most a quantity including, inter alia, squared terms of the norm of ? might be involved. The board can see no obvious reason why the skilled person would then take the square root of these terms only in order to obtain the norm of ?, and neither did the respondent provide one. In the board's view, the respondent's attack clearly relies on a highly speculative ex-post facto analysis.

7.6 The board concludes that the subject-matter of claim 1 involves an inventive step when starting out from D12 (Articles 52(1) and 56 EPC).

8. First auxiliary request - inventive step - the combination of D1, D9 and D12

The respondent also argued that the skilled person would arrive at the subject-matter of claim 1 by combining the documents D1, D9 and D12. However, since D12 does not disclose the inventive feature of basing the hearing aid gain on a norm of the vector of coefficients, even combining D12 with D1 and D9 would not lead to the subject-matter of claim 1. Consequently, this attack also fails.

9. Dependent claims

No objections were raised concerning the dependent claims, i.e. claims 2 to 9.

10. Grounds for opposition under Article 100(b) and Article 100(c) EPC

The opposition division found that the grounds for opposition pursuant to Article 100(b) and (c) EPC (cf. Articles 83 and 123(2) EPC) did not prejudice the maintenance of the patent as granted. The respondent has not contested this part of the decision and did not raise any objections based on these grounds in respect of the present first auxiliary request. Nor does the board see any reason to raise any such objections.

11. Conclusion

The claims according to the first auxiliary request meet the requirements of the EPC. The patent can therefore be maintained in amended form on the basis of this request (Article 101(3)(a) EPC). However, the board has not examined the description or drawings for compliance with the EPC, and considers that this matter is best dealt with by the opposition division.