In a hearing instrument, feedback refers to an unstable condition that typically results in a high-pitched whistling sound generated by the device. Feedback, in electroacoustic terms, is defined as the return of the output of a machine, system or process to the input. The path from the system’s output to the system’s input is called the feedback path. The system becomes unstable when the feedback is strong enough to drive the system by overpowering the normal input, causing it to behave like an oscillator.1

Hearing instrument feedback can be both annoying and distracting to the wearer, and it can reduce the effectiveness of the device. Not only does the whistling reduce audibility, but the most common solution—lowering the gain— can result in a further reduction in audibility. Hearing instrument whistling can also be a social embarrassment for the wearer if it is loud enough to be audible to others nearby.

There are two major factors that influence feedback: gain and insufficient attenuation between the hearing instrument microphone and receiver, both of which are addressed below.

Gain and Feedback
The most important factor affecting the onset of feedback is the gain setting of the device. The higher the gain, the greater the likelihood for feedback to occur. This relationship makes multi-band wide dynamic range compression (MBWDRC) hearing instruments, which employ high gain settings at low input levels, more prone to generating feedback than linear hearing instruments with uniform gain settings at all signal levels.

The simplest way to deal with feedback is to reduce the hearing instrument gain until the feedback subsides. However, reducing the gain limits the maximum amplification the device can deliver. In many instances, the dispensing professional would like to prescribe more gain but cannot because the hearing aid would start to whistle.

Feedback Path Attenuation
If there is insufficient attenuation in the path from the hearing aid receiver to the microphone (feedback path), feedback will occur at comparatively low gain settings. Often, a large vent is part of the feedback path. In such cases, the attenuation of the feedback path can be increased by simply plugging the vent or using a smaller vent size.

The main factors that affect the attenuation of the feedback path are:

Acoustic environment near the ear: Reflecting surfaces near the ear, such as a telephone handset or proximity of another person’s head during a hug, can temporarily reduce feedback path attenuation by guiding a larger portion of the output signal back to the hearing aid microphone.

Tightness of the fitting (leakage around ear molds): A loose fit can create a more efficient path for the acoustic energy in the ear canal to feed back to the hearing aid microphone (i.e., there is less attenuation between the receiver and the microphone).

Geometric configuration: Feedback is more problematic in CIC and ITC devices than in BTE devices because, in the former, there is less separation between the receiver and the microphone, which results in less attenuation between the receiver and the microphone.

Vent size: Although they provide many advantages (such as direct sound and reduced occlusion effect), large vents generally decrease the attenuation between receiver and microphone and therefore increase susceptibility to feedback.

Conventional Feedback Management Techniques
Good audiological practice is perhaps the most effective approach to controlling feedback. Elements of this practice include:

• Taking an excellent impression and ensuring a tight fit of the earmold;
• Choosing appropriate vent sizes; and,
• Selecting a device that is appropriate for the amount of amplification needed.

These techniques can be effective in attenuating the feedback path and should be used whether or not advanced feedback management tools are available.

Feedback Suppression Through Gain Reduction
If conventional techniques are insufficient to prevent feedback at the desired gain, the gain must be reduced to a level at which feedback does not occur. This forces the dispenser to compromise the fitting with a gain limitation that may result in less-than-optimal audibility. This compromise has the greatest negative impact on patients with severe-to-profound hearing losses since they require a greater degree of amplification.

To minimize the detrimental effects of gain limitation, some hearing instrument manufacturers use one or several notch filters to reduce gain in a narrow frequency range in the region of the feedback frequency. However, even with the narrowest of notch filters, the frequency response of the device can be compromised. Secondly, there is partial reduction in gain for frequencies adjacent to the band blocked by the notch filter. Moreover, intermittent failure of feedback suppression is more likely to occur with narrow notch filters as changes in the wearer’s acoustic environment can easily shift the feedback frequency out of the notch filter’s narrow suppression band. To achieve stable feedback suppression, a wider notch filter is generally necessary. Unfortunately, the wider the frequency band over which the notch filter reduces gain, the greater the loss in signal audibility and speech intelligibility.

An alternative approach to feedback suppression is to compensate for moment-to-moment fluctuation in the feedback frequency by using adaptive notch filters that continually sample the feedback and alter the notch settings accordingly. While this approach makes feedback reduction more reliable, it does not solve the problem of reduced audibility due to lower gain.

Fig. 1. Block diagram of the operation of the Digital Feedback Suppression algorithm.

Suppression of the Feedback Signal
With Digital Feedback Suppression (DFS), the hearing instrument estimates the attenuation of the feedback path as a function of frequency, then creates a digital filter with the same response. This digital filter is then applied to the signal-processing path in parallel with the actual feedback path. Note that, at the point where the digital filter model of the feedback path and the actual feedback path meet, the two paths are subtracted (Fig. 1). This has the effect of suppressing (or compensating for) the feedback path. Ideally, this approach eliminates the feedback signal without altering the input signal, because the signal path remains unaffected.

Fig. 2. The feedback path is first measured by comparing the noise transmitted by the receiver to the noise picked up by the microphone. The signal path through the hearing instrument is disabled during this measurement.

For the feedback-path model to work, the feedback path must first be measured. This measurement is made during the initial fitting session (Fig. 2). It is important that the measured feedback path be representative of the in-situ use of the device. It is sometimes mistakenly thought that feedback must actually be occurring for this measurement to be performed. This is not the case. In fact, because the signal path through the device is disabled during the measurement, the device will not whistle during the measurement.

It is also interesting to note that it is easier to perform this measurement with a larger vent, because less physical attenuation in the feedback path makes for a stronger test signal and a more reliable measurement. This results in a better estimate of the feedback path and hence better performance of the DFS system. Therefore, one immediate benefit of this system is the ability to use larger vent sizes.

Adaptive Digital Responses
If the feedback path remains constant, the system will perform as expected and the feedback will be cancelled. The result will be an increase in headroom. The headroom increase is the amount by which the feedback-free gain can be increased with DFS active above the maximal feedback-free gain with inactive DFS.

Fig. 3. Change in the feedback path with changing acoustic condition. Blue: Response of feedback path of a hearing instrument (in-situ) for a woman wearing her hair down. Red: Response of feedback path for the same listener wearing her hair up.

However, DFS cannot eliminate feedback under all conditions and circumstances. In the course of a day, many actions, such as bringing a hand to the ear or putting on a hat, can alter the feedback path so that it no longer matches the DFS filter. To retain effective feedback cancellation in changing acoustic environments, the hearing instrument needs to continually measure the feedback path and update the DFS filter accordingly (Fig. 3).

To measure the feedback path while the hearing instrument is in use, some feedback suppression algorithms analyze a low-level noise introduced into the receiver while the hearing instrument is in use. This may be a less-than-optimal method of feedback reduction, because the user experiences random noise signals which decrease the signal-to-noise ratio (SNR), and these signals have the potential to impair both signal audibility and speech intelligibility.

Kates^1,2 describes an alternative method for estimating the feedback path without introducing a noise. This approach is used in the GN ReSound Canta, Digital 5000 and Danalogic hearing instruments. Instead of introducing a noise, this algorithm uses ambient sounds in the environment to estimate and minimize the difference between the actual feedback path and the feedback path modeled by the device. This adaptive system is designed to continually update the feedback estimation relative to transient changes in the acoustic environment.

During operation, the system can occasionally produce an artifact if the input signal contains a persistent sinusoid pattern, such as a microwave beep or a sustained tone from a musical instrument. In these situations, the algorithm may interpret the sinusoid as feedback and attempt to cancel it. Experience has shown that when properly counseled, wearers are willing to tolerate such artifacts in exchange for the increased audibility provided by adaptive DFS. In addition, the device is programmable, and wearers always have the option of employing a program with slow/non-adaptive DFS or with DFS disabled.

Measuring DFS Benefit
Feedback problems are common: out of 383 hearing instrument wearers tested in clinical studies for GN ReSound, 48% experienced feedback with their preferred gain setting. For 73% of those, the preferred gain setting could be achieved through DFS.

Fig. 4. Increase in feedback-free gain (headroom) with active DFS in 10 ITE users. Test-retest reliability is better than 1 dB.

To verify the increase in usable gain afforded by this feedback suppression system, 10 users were fitted with ITE devices. For each subject, DFS was turned off and the gain was increased until feedback occurred. The system was then activated, which suppressed the feedback, and the gain was increased until feedback reoccurred. The difference in gain at which feedback occurred with and without the system (headroom) is shown in Fig. 4.

Fig. 5. Mean 2cc coupler gain increases for 10 listeners after activation of DFS. Gain was increased until the user’s preferred gain setting was reached or until the device became unstable (feedback occurred), whichever gain setting was lower.

Fig. 5 illustrates to what extent the additional gain is actually used. These results are in agreement with earlier findings by the Central Institute for the Deaf that feedback does indeed limit the gain prescribed to patients.³

In a GN ReSound in-house study, five BTE and five ITE users with feedback problems at their preferred gain settings were fitted with three programs:

1. Multi-band Wide Dynamic range compression (WDRC);
2. WDRC and DFS; and,
3. WDRC, DFS and Noise Reduction (which reduces gain for unmodulated sounds in the environment).

Fig. 6. Subjective evaluation of three programs by 10 listeners, with the number of subjects that preferred a particular program in three listening situations and their judgement of overall preference.

Fig. 6 shows the results of a subjective evaluation of the programs by the 10 listeners. To emphasize the effect of DFS, preferences for programs 2 and 3 were grouped together. The listeners appear to prefer programs with DFS, because it permits feedback-free gain settings appropriate for their hearing loss.

Improvement in speech intelligibility is an important measure in evaluating the benefit provided by a hearing instrument. Fig. 7 shows the presentation level of speech in quiet that is required to correctly identify 50% of the sentences in the Hearing In Noise Test (HINT)⁴ for fittings with and without DFS. When the system is not active, the hearing instrument cannot provide as much gain as when it is active and the speech level must be raised to sustain 50% intelligibility. The data show improved intelligibility with active DFS in eight of the ten subjects. For six of the eight, the improvement is statistically significant. Two subjects performed worse with active DFS.

Fig. 7. Performance in HINT test for listeners evaluated in Fig. 6. Stars indicate significant performance differences.

Conclusion
Digital Feedback Suppression is one example of how digital hearing instrument technology enables solutions to long-standing problems in amplification. It is unlikely that signal processing of this complexity can be achieved with analog devices.⁵ Compared with other methods of feedback reduction, such as gain limiting and/or notch filtering, the digital system is designed to suppress feedback without sacrificing audibility of any portion of the signal.

The system may be especially helpful in ITC and CIC hearing instruments, where the microphone and receiver are in close proximity. DFS also enables the use of larger vent sizes, reducing the occlusion effect. In hearing instruments utilizing WDRC, the system may be particularly effective in facilitating feedback-free performance because these devices are designed to provide high gain at low input levels, and therefore are prone to generate feedback.

This article was submitted to HR by Laurel Olson, MA, director of clinical audiology, Hannes Müsch, PhD, research engineer, and Christopher Struck, BSEE, manager of eletroacoustics, of GN ReSound. Correspondence can be addressed to HR or Laurel Olson, GN ReSound, 220 Saginaw Dr., Redwood City, CA 94063; email: lolson@gnresound.com.

References
1. Kates J: Constrained adaptation for feedback cancellation in hearing aids. J Acoust Soc Amer 1999; 106: 1010-19 and references therein.

2. U.S. Patent Office: U.S. Patents #6,072,884 and #6,219,427.

3. St-French GM, Wood DJ & Engebretson AM: Behavioral assessment of adaptive feedback equalization in a digital hearing aid. J Rehab Res Dev 1993; 30 (1): 17-25.

4. Nilsson M, Soli SD & Sullivan JA: Development of the Hearing In Noise Test for the measurement of speech reception thresholds in quiet and in noise. J Acoust Soc Amer 1994; 95 (2): 1085-99.

5. Groth J: Digital signal processing has made active feedback suppression a reality. Hear Jour 1999; 52 (5): 32-36.