Note: This is a continuation of Adaptive Feedback from a previous blog on Adaptive Feedback V – Solution #9.
Adaptive Feedback Cancellation History
The literature is filled with varying approaches to adaptive feedback cancellation. Any attempt to include all of the approaches will fall short, and this blog is no exception. Still, some of the techniques and time-line are provided below, including both non-continuous and continuous adaptive feedback techniques. Noteworthy is the fact that in the early-to-mid 1980s, the work by Nunley, et. al. (1983) {{1}}[[1]] Nunley J, Staab W, Steadman J, Wechsler P and Spencer B. 1983. A wearable digital hearing aid, The Hearing Journal, Vol. 36, 29-31 [[1]] and by Engebretson, et. al. (1985), {{2}}[[2]] Engebretson, A.M., Morley, R.E., and Popelka, G.R. Hearing aids, signal supplying apparatus, systems for compensating hearing deficiencies, and methods. US Patent No. 4,548,082, 1985 [[2]] which were among the first to use digital signal processing (DSP) in a wearable, but non-commercial hearing aid, actually employed continuous adaptive feedback cancellation with the use of the LMS filtering and a two-microphone system (Figure 22). However, it was not until almost 10 years later, when digital signal processing was made available in commercial hearing aids, that continuous adaptive signal processing was revisited. The intervening time period used “adaptive” feedback cancellation techniques that were of the non-continuous nature.
Figure 22. Schematic of one of the experimental two microphone input adaptive digital LMS filter systems used for background noise and feedback reduction (Courtesy of Audiotone, 1983 – Redrawn from original).
Probe Signal
Another classical approach to decorrelation (Acoustic Feedback Solution #9 – Adaptive Feedback)has been the addition of a probe signal to the speaker input to measure and control the adaptation of the feedback canceller as shown in Figure 23. {{1}}[[1]] Sandlin, R.E. 1996. Introducing a completely digital hearing instrument. The Hearing Journal, Vol. 49, No. 4, April, p. 48.[[1]] A number of different adaptive feedback methods have used, and continue to use a probe procedure, often referred to as a feedback manager test, to determine the feedback path.
Figure 23. The use of a probe signal, identified as a composite signal generator and earmold analyzer. Sandlin, 1996.
General Adaptive Feedback Cancellation-Based Feedback Reduction Approaches
Adaptive feedback reduction/cancellation algorithms essentially require digital technology, even those digital/analog hybrids that were able to provide some very limited active suppression. For those that use DSP (digital signal processing) to reduce feedback, there are two primary approaches, both of which use continuous adaptation methodology:
- Those that act on the feedback signal when the hearing aid goes into oscillation and generates a howling/whistling sound, and
- Those that act on the transfer function, which is related to the frequency-amplitude-phase relationship of the signal in the feedback path, even prior to the hearing aid generating this acoustic feedback. These generally used a probe signal to measure the transfer function as shown in Figure 24.
Related to the #2 approach, an early Danavox analog/digital hybrid power hearing aid used a fairly loud broadband test signal emitted into the signal path in the ear to determine the transfer function of the feedback path. {{3}}[[3]] Algorithms. Feedback: “search and destroy” feedback suppression has arrived. 2004. Danavox Danalogic Company Brochure.[[3]] This information was used to adjust a filter in the DSP chip portion of the hybrid hearing aid in such a way that the feedback signal was cancelled whenever the feedback conditions were met. Unfortunately, the test signal was obtrusive to users and to those around them.
The literature is filled with a plethora of feedback reduction signal processing approaches. A few of the primary approaches that have been used, follow. All are based on the use of digital signal processing. However, it is not possible in a blog such as this to cover all of the various feedback reduction signal processing approaches that have been used. The following identification of approaches is merely an attempt to identify some of the techniques that have been used, with the realization that many of the principles are used in some of the latest systems, which are not included.
Category 1. Systems Requiring a Feedback Path Measurement During Fitting (Initialization)
These hearing aids utilize a feedback path measurement to be made during fitting. This determination of the feedback settings is called initialization (some call this a feedback manager test), and is performed during the fitting with the hearing instrument being worn. This feedback path is used as a starting point for the adaptive feedback canceller. These use a signal probe inserted into the output signal path of the feedback suppression circuit to supply a source of feedback during times of little containment of the undesired feedback signal being present within the audio environment of the circuit. Such a system, which first determines the feedback path transfer function and then calculates filter coefficients to cancel the feedback, is shown in a general overview in Figure 24 by Soli, et. al., 1995. {{4}}[[4]] Soli, S.D., Buckley, K.M., Widin, G.P. 1995. Auditory prosthesis, noise suppression apparatus and feedback suppression apparatus having focused adaptive filtering. US Patent No. 5,402,496. [[4]] They describe a noise and feedback suppression apparatus to process audio input having both a desired component and an undesired component that uses a probe (test signal) into a hearing aid output open loop.
Figure 24. Use of a probe test signal to identify and model a feedback suppression hearing aid.
Adaptive Feedback Reduction Based on Gain Manipulation
These are sometimes called feedback management systems. The gain is reduced in the frequency channel(s) in which feedback occurs, but only when feedback is detected in that channel using a probe test signal. The amount of gain reduction is variable – by channel, by frequency, and by engineering design – depending on the desired output result. These systems use a feedback manager test that is conducted in the initial hearing aid fitting session. This feedback manager test is also identified as “initialization.” During the test, gain is set to maximum and the feedback detector to the probe signal monitors the presence of tonal and periodic signals. If no tonal signal is detected in a channel, maximum gain is allowed. However, if a tonal or whistling feedback signal is detected, the feedback manager automatically reduces the gain in that channel until feedback disappears and sets this level as the maximum allowable gain in that channel. Keep in mind that this gain level is based on the in-situ fitting. To manage abrupt changes in the acoustical environment, an adaptive gain component is involved. If feedback is detected in any of the channels resulting from environmental sounds, gains of the frequency channels are reduced by the same mechanism used in the feedback manager test. An example of this would be the Phonak Perseo hearing aid.
Adaptive Feedback Reduction Based on Notch Filtering
While the use of notch filtering has been presented in the feedforward and fixed-feature feedback management section (Hearing Aid Acoustic Feedback III), adaptive notch filtering is used to generate multiple sharp notch filters at feedback frequencies during the feedback manager test, using a probe signal, and employ an adaptive component that is activated by abrupt changes in the acoustic environment after the feedback manager test has been completed – during normal use of the hearing aid. For abrupt changes, those filters not activated by the feedback manager test are used. The notch filters by different hearing aid manufacturers may differ in the number of notch filters, their depth, sharpness, and operating frequency range. An example of this would be the Siemens Triano hearing aid.
Adaptive Feedback Reduction Based on Phase Cancellation
These systems use algorithms that use phase cancellation to monitor the transfer function of the feedback path, generate a signal with a transfer function that is similar to the feedback path, and then subtract the generated signal from the hearing aid microphone output. In other words, they mimic the feedback path in frequency-amplitude-phase. These are sometimes referred to as feedback cancellation hearing aids.
Early versions estimated the transfer function of the feedback path by injecting a low-level noise into the receiver and then used a cross-correlation approach of the receiver to the microphone input. The digital filter coefficients were modified to create a phase cancellation signal. However, the continuous, low-level noise was found by many hearing aid users to be annoying {{5}}[[5]] Algorithms. Feedback: “search and destroy” feedback suppression has arrived. 2004. Danavox Danalogic Company Brochure [[5]].
Later versions eliminated the noise injection with a noise burst during the initial fitting session, and reduced acoustic feedback by using multiple signal processing stages. Fixed and adaptive filters were used. A fixed filter transfer function was determined in a feedback path estimation using a feedback manager test during the initial in-situ fitting session. A noise burst was sent to the receiver and cross-correlated with the microphone output. Some systems use multiple delay times to estimate the transfer functions of multiple feedback paths when the hearing aid is worn. In some hearing aids an adaptive filter is implemented to act on the dynamic components of feedback. This uses the fixed filter as a baseline but then performs a constant cross-correlation between the receiver input and the microphone output to monitor changes in the transfer function. If a change is detected, the algorithm changes the characteristics of the adaptive filter to approximate the transfer function of the new feedback signal path. A signal with the combined fixed and adaptive filters is generated internally in the hearing aid and subtracted from the microphone output to cancel feedback.
As a more detailed example, a broadband test signal approximately 10 seconds in duration is emitted by the hearing aid as the probe signal. To enable this procedure, specialized signal processor software is downloaded to the hearing aid. Once the test is completed, the default hearing aid functions are restored. The test signal is used to determine the coefficients of a fixed filter that mirrors the static feedback path. This part of the system ensures that feedback caused by factors such as venting or sound leakage around an earmold will not occur. A second element is an adaptive filter that continuously adapts to the current feedback conditions. The input signal is monitored, and this filter is able to change its characteristics in order to cancel feedback in situations which would not arise when the wearer is sitting quietly. The feedback compensation signal is equivalent to the external feedback signal, but of opposite phase. An example of this would be the ReSound Canta hearing aid.
In another form, when used in conjunction with a slow adaptation constant, the canceller’s behavior is similar to using constrained adaptation. This means that temporarily large variationsfrom the measured feedback path will not be tracked.
Some systems use algorithms that function in either the slow or fast mode rather than using two adaptive digital filters. Two detectors are used – one for howl and the other for feedback change, and used primarily for high frequencies above approximately 1500 Hz to avoid canceling low-frequency tonal sounds. These can be used also with the feedback manager test if the feedback cancellation algorithm is unable to suppress the acoustic feedback {{6}}[[1]] Oticon. 2004. The Syncro Audiological Concept.[[6]]
The use of combining different strategies has also been used in the Widex Senso Diva {{7}}[[7]] Kuk, F. and Ludvigsen, C. 2002. The real-world benefits and limitations of active digital feedback cancellation. Hearing Review 9(4):64-68.[[7]] This system used both phase cancellation and adaptive gain reduction to control feedback. The adaptive portion incorporated a fixed and a slow-acting adaptive filter (utilizing phase cancellation), and a fast-acting gain reduction feature.
Another strategy combination involved near and far feedback loops {{8}}[[8]] Wang, R., and Harjani, R. 1993. Acoustics, speech, and signal processing, ICASSP-93, 1993 IEEE International Conference on…, Minneapolis, MN.[[8]] They developed a feedback cancellation scheme that manages two different types of oscillation: a first open-loop feedback managed by phase cancellation, and a second feedback related to an open-loop transfer function outside the primary audio frequency region managed by negative feedback. Both of the feedback procedures manage what are recognized as near and far feedback loops which are required for a hearing aid to provide maximum usable gain of the overall hearing aid system.
Still another variation might take the feedback path and combine it with a frequency-dependent gain limiter with the gain limiter applied at all times, even when the feedback canceller is disabled. This approach is used to limit gain above the MSG (maximum stable gain) when the feedback canceller is disabled to prevent feedback occurring in situations when an object approaches the ear. Other variations take this feedback path alone and use it as the adaptive feedback canceller.
Category 2. Feedback Cancellers Not Requiring Initialization of the Feedback Path
These approaches use the measured feedback path as a starting point for the adaptive feedback canceller. The adaptation speed of the feedback canceller is selected based on an analysis of the input signal. These might be called feedback cancellers with automatic speed control. These will be discussed in a future blog.
More Recent Direction at Acoustic Feedback Management in Hearing Aids
Hearing-impaired individuals generally utilize hearing aids binaurally. It would appear, therefore, that feedback cancellation systems might be improved, especially against sinusoidal or narrowband input signals by systems where the hearing aids on each side of the head interact with each other. One of the basic ideas is that oscillations detected by one hearing aid can only be caused by feedback if the aid on the other side did not detect oscillations of exactly the same frequency. This binaural oscillation detection approach makes use of the head shadow effect and requires a data link between the two hearing aids and is diagrammed in Figure 25.
Figure 25. Binaural oscillation detection for feedback suppression.
Adaptive feedback cancellers, along with more recent approaches to acoustic feedback solutions, as described briefly in the next paragraph, will have to await a future series of blogs on adaptive feedback cancellation in hearing aids.
