Acoustic factors affecting interaural level differences for cochlear-implant users

Abstract

Bilateral cochlear-implant (BICI) listeners primarily use interaural level differences (ILDs) to localize sound in the horizontal plane. However, the ILD magnitude is altered at different frequencies and azimuths due to a combination of several acoustic phenomena such as the acoustical bright spot, acoustic axis, and microphone porting. This paper investigated the effects of BICI microphone placement on ILDs through an analysis of head-related transfer functions. At-the-canal BICI microphone placement provided both larger and more monotonic ILD-azimuth functions than behind-the-ear microphone placement. Results have implications for the fitting of clinical devices and their effect on sound localization in BICI users.

1. Introduction

Interaural time differences (ITDs) and interaural level differences (ILDs) are used to perform horizontal-plane or azimuthal sound localization. For acoustic-hearing listeners, there is a dominant role of low-frequency (<1.5 kHz) ITDs in azimuthal localization (Wightman and Kistler, 1992). In contrast, for bilateral cochlear-implant (BICI) listeners, there is a dominant role of ILDs (e.g., Aronoff et al., 2010). This is due to the envelope-based signal processing in modern cochlear-implant (CIs), which use bilaterally unsynchronized processors that discard low-frequency temporal fine structure in exchange for rapid sampling of the amplitude envelope. While two implants provide more accurate localization than one, neither case provides spatial-hearing abilities comparable to acoustic-hearing listeners (e.g., van Hoesel and Tyler, 2003).

In acoustic hearing, the pinna and ear canal collect sound to further excite the peripheral auditory system. In contrast, CIs collect sound using microphones typically placed on a behind-the-ear (BTE) processor. There are alternative placements such as microphones located at an off-the-ear processor, as in the case for single-unit processors with microphones located at the side of the head, or accessory microphones near the opening of the ear canal. At-the-canal (ATC) microphone placement has been shown to improve speech understanding in noise and sound localization for BICI users (Kolberg et al., 2015; Mantokoudis et al., 2011; Jones et al., 2016). Though the referenced studies used prior generation technology with highly recessed BTE microphones, the reason for these improvements is unclear; it could be related to the physical acoustics of different microphone placements. This paper investigates the effects of microphone placement on the primary cue for azimuthal sound localization in BICIs. ILDs for BICI microphone placements were analyzed as functions of frequency and azimuth, and contrasted with the ILDs delivered to acoustic-hearing listeners.

ILDs as a function of azimuth are highly frequency dependent. The head is an approximate spherical obstructer of sound, and sound originating off the midline must diffract around the head to reach the ear farther from the source. A sound will diffract around a head at a specific frequency if its wavelength is larger than the head diameter, approximately 24.1 cm for humans (Lee et al., 2006). Relatively short wavelengths compared to the head diameter are more likely to be reflected or absorbed. Therefore, the sound level at the ear farther from the source becomes progressively attenuated as frequency increases, creating larger ILDs for larger azimuths.

However, according to Fresnel's theory of diffraction, a wave obstructed by a sphere will cause a “bright spot” on the side opposite to the incident wave. This bright spot occurs due to the incident wave diffracting around the sphere and arriving at the opposite side in phase, causing constructive interference.

In acoustics, the Fresnel bright spot is known as the “acoustical bright spot,” and it results in highly non-monotonic ILD-azimuth functions. Non-monotonicities of this kind are perceptually relevant and result in the mislocalization of 1.5-kHz tones in normal-hearing listeners (Macaulay et al., 2010). However, the non-monotonicities disappear at higher frequencies for mannequin measurements. For example, acoustic mannequin measurements found that the ILDs at 6 kHz for 30°, 60°, and 90° azimuth are approximately 14, 20, and 30 dB, respectively (Kayser et al., 2009). In contrast, a spherical head model without facial prominences shows much smaller ILDs of 8, 13, and 6 dB, respectively (Kayser et al., 2009). The exact cause of these differences is unclear, but a main distinction between a sphere and acoustic mannequin is the addition of the pinnae. The pinnae should obstruct high-frequency soundwaves traveling around the back of the head from entering the ear canal, causing them to not interact with soundwaves traveling around the front. This would result in a reduced bright spot effect at high frequencies, thus re-establishing ILDs as a robust cue for azimuthal sound localization.

Constructive interference from the acoustical bright spot increases gain at the ear farther from the source (i.e., the contralateral ear). However, the acoustical bright spot is not the only contributor to non-monotonic ILD-azimuth functions; there are other factors that may affect gain at the ear closer to the source (the ipsilateral ear).

The pinna acts as a directional acoustic reflector and amplifies high-frequency sounds as they enter the ear canal. The amount of gain provided by the pinna is not constant across azimuth resulting in non-monotonic monaural gain-azimuth functions. The source azimuth that creates the highest ipsilateral gain is known as the “acoustic axis” (e.g., Middlebrooks and Pettigrew, 1981). The azimuth and corresponding gain of the acoustic axis may be affected by microphone placement and influence the non-monotonicity of ILD-azimuth functions. Additionally, head reflections or multiple microphone ports may cause comb filtering, or periodic dips in the frequency spectrum from summed phase delays, resulting in reduced monaural gain at specific azimuths and frequencies.

In summary, constructive interference from the acoustical bright spot affects contralateral gain. Factors such as the acoustic axis, head reflections, and microphone porting affect ipsilateral gain. Based on a combination of these effects, certain CI processor microphone placements may convey more monotonic ILD-azimuth functions and thus be better suited for azimuthal sound localization. This study compared ILDs as a function of frequency and azimuth for different BICI microphone locations, and examined the microphones' ipsilateral and contralateral gains to better understand the effects of the bright spot, acoustic axis, and microphone porting. Considering the improved localization abilities seen with ATC microphones, we hypothesized that ATC microphone placement would provide more monotonic ILD-azimuth functions.

2. Analysis

2.1. Methods

An ILD analysis was conducted on five head-related transfer function (HRTF) microphone placement libraries recorded using head and torso simulators. ILDs were measured at several frequencies and azimuths for each library, and the resulting ILD-frequency-azimuth functions were compared between microphone placements. ILDs were averaged within 1 equivalent rectangular bandwidth (Moore and Glasberg, 1983) at center frequencies of 0.25, 0.5, 0.75, 1, 1.5, 2, 3, 4, 5, 6, 7, and 8 kHz. Bands were separated using rectangular Fourier filters. ILDs were calculated as the right/left ratio of the root-mean-square amplitude converted to dB. Ipsilateral and contralateral gains were calculated with the same parameters using the root-mean-square gain relative to 0° azimuth.

The in-the-canal (ITC) and BTE microphone measurements were published by Kayser et al. (2009) and recorded on a Bruël & Kjær (Nærum, Denmark) Type 4128 C mannequin at a 0.8 m loudspeaker distance in an anechoic chamber at 5° azimuthal increments. The ITC library was recorded using the mannequin's in-ear microphones. The BTE library was recorded using the single-port front microphone of a Siemens (Munich, Germany) Acuris BTE hearing aid with no directional processing. The BICI HRTF microphone placement recordings were created by Advanced Bionics (Valencia, CA) using the front BTE and ATC (T-Mic®) microphones on a Naída CI Q90 sound processor. Each BICI microphone was measured on a G.R.A.S. (Holte, Denmark) KEMAR in 10° increments, with measurements made approximately 1.4 m from the loudspeaker in an anechoic environment. The Q90 front BTE microphones have two ports, one on either side of the processor, and will be distinguished as “BTE2.” In addition to the explicitly recorded libraries, an additional set of BICI HRTFs was formed by taking a 50/50 linear blend of the BTE2 and ATC head-related impulse responses prior to filtering. This blend is the default clinical setting when a T-Mic® is connected to the processor. Including these measures may highlight the repercussions of leaving this parameter unchanged (as is frequently done in clinical settings) (Kolberg et al., 2015).

2.2. Results

Figure 1(A) shows the calculated ILD-azimuth functions of the different microphone placements for each frequency band. ILD-azimuth functions after 30° for the different microphone placements diverge by 8 kHz, with differences as large as 30 dB at some azimuths. A three-factor analysis of variance (ANOVA) on the ILD-frequency-azimuth data found main effects of microphone placement [F_4,584 = 17.44, p < 0.001, η² = 0.06], larger ILDs for high (>4 kHz) vs low (≤4 kHz) frequencies [F_1,584 = 248.11, p < 0.001, η² = 0.20], and larger ILDs at large (≥50°) vs small (<50°) azimuths [F_1,584 = 260.11, p < 0.001, η² = 0.21]. The interactions of placement × frequency [F_4,584 = 20.64, p < 0.001, η² = 0.07], placement × azimuth [F_4,584 = 6.79, p < 0.001, η² = 0.02], and azimuth × frequency [F_1,584 = 30.95, p < 0.001, η² = 0.03] were all significant. The three-way interaction was also significant [F_4,584 = 10.96, p < 0.001, η² = 0.03], corroborating the clear effect of microphone placement on ILDs for high frequencies and large azimuths. Tukey's honestly significant difference (HSD) post hoc comparisons found that the ITC and ATC microphone placements had significantly larger mean ILDs than all other microphone placements (p < 0.05). However, the ITE and ATC libraries were not significantly different from each other (p > 0.05).

Fig. 1.

(Color online) (A) ILD-azimuth functions of microphone placements measured in different frequency bands. For each subfigure, the x axis is clockwise azimuth (degrees) and the y axis is ILD (dB). (B) Ipsilateral and contralateral gain-azimuth functions for the microphone placements. To avoid visual clutter, only functions for the BTE, BTE2, and ATC microphone placements are shown. For each subfigure, the x axis is clockwise azimuth (degrees) and the y axis is gain relative to 0° azimuth (dB).

Considerable non-monotonicities in the ILD-azimuth functions can be seen for frequencies as low as 1.5 kHz, with reversals visible for all microphone placements. However, Fig. 1(B) reveals that the mechanisms behind these non-monotonicities are not the same. For example, below 5 kHz, the non-monotonicity of the BTE ILD-azimuth function appears to be driven by variability in contralateral gain, while the BTE2 ILD-azimuth function is primarily driven by a large decrease in ipsilateral gain.

Two measures were developed to quantify non-monotonicities in the ILD-azimuth functions. For the first measure, we subtracted the ILD at 90° from the largest ILD of a given ILD-azimuth function and expressed the difference as a percentage of the largest ILD. This measure was developed because it describes large reversals in the ILD-azimuth functions and takes into account the range of available ILDs. The resulting percentages can be seen in Fig. 2(A). A two-way ANOVA found a main effect of microphone placement [F_4,50 = 2.66, p < 0.05, η² = 0.15] on the percentages and no main effect of high (>4 kHz) vs low (≤4 kHz) frequencies [F_1,50  =  0.12, p > 0.05, η² < 0.01]. Critically, the two-way interaction was significant [F_4,50 = 3.95, p < 0.01, η² = 0.22]. Tukey HSD post hoc comparisons within the high-frequency data found the ITC microphone placement significantly more monotonic (p < 0.05) than the BTE2, but not the BTE, 50/50, or ATC (p > 0.05 for all). The ATC was also more monotonic than the BTE2 (p < 0.05), but not the BTE or 50/50 (p > 0.05 for all). The BTE, BTE2, and 50/50 were not significantly different from each other (p > 0.05 for all).

Fig. 2.

(Color online) (A) Difference between the largest ILD and the ILD at 90° for each library as a function of band center frequency and expressed in percent of the maximum ILD for each function. (B) The proportion of azimuths where an ILD-azimuth function is below the previous local maximum as a function of band center frequency.

For the second measure, we counted the number of azimuths where the corresponding ILD is less than the preceding local maximum and expressed the number as a percentage of the total number of azimuths. This measure was developed to describe small reversals in the ILD-azimuth functions which may result in azimuthal mislocalization. The resulting percentages can be seen in Fig. 2(B). A two-way ANOVA also found a main effect of microphone placement [F_4,50 = 3.4, p < 0.05, η ²= 0.20] on non-monotonicity, no main effect of high (≥5 kHz) vs low (≤4 kHz) frequencies [F_1,50 = 0.58, p > 0.05, η² = 0.01] or no significant interaction [F_4,50 = 1.25, p > 0.05, η² = 0.07]. Tukey HSD post hoc comparison found the ATC placement significantly more monotonic than the BTE2 (p < 0.05). No other comparisons were significant (p > 0.05).

3. Discussion

We hypothesized that ILD-azimuth functions would be more monotonic for ITC and ATC compared to BTE microphone placements. The results in Figs. 1 and 2 generally support the main hypothesis and suggest that ITC and ATC microphone placements provide more monotonic ILD-azimuth functions. While the tested microphone placements have similar ILD-azimuth functions between 0.25 and 1 kHz, they diverge starting around 1.5 kHz [see Fig. 1(A)]. This is not surprising, as the wavelength of 1.5 kHz (22.9 cm) is approximately the size of the average human head (24.1 cm or 1.43 kHz). Shorter wavelengths (i.e., higher frequencies) should be increasingly obstructed by the head and give rise to larger ILDs, and it is above this frequency that the differences in BICI microphone placements becomes clear.

Above 2 kHz, Fig. 1(A) shows that the ITC and ATC microphone placements consistently have the largest ILDs, especially for azimuths ≥50°. At about 5 kHz and above, the BTE, BTE2, and 50/50 BTE2/ATC microphone placement blends have significantly smaller ILDs for nearly all azimuths. For microphone placements that should have a relatively large contribution of pinna filtering (ITC and ATC), ILD-azimuth functions become relatively monotonic at 5 kHz. As seen in Fig. 2, all microphone placements become more monotonic between 2 and 5 kHz. However, the functions have varying degrees of monotonicity above 5 kHz with the ITC, 50/50, and ATC microphone placements being noticeably more monotonic and the BTE and BTE2 microphone placements being largely non-monotonic. The 5-kHz cutoff is interesting because the wavelength of 5 kHz (6.89 cm) is approximately the height of the average adult pinna (6.45 cm), implying these results are due to the acoustical bright spot (Kalcioglu et al., 2003).

However, the results from Fig. 1(B) reveal that the mechanisms behind these non-monotonicities are not the same. The BTE placement has a highly variable contralateral gain, presumably due to the bright spot, while the BTE2 placement has a variable ipsilateral gain. The BTE2 microphones each have two ports, one on either side of the processor. At 0° azimuth, sound enters both ports at the same time. As we approach 90° azimuth, sound enters the ports at different times. This acoustic delay may result in a comb filter and decreased ipsilateral gain for high frequencies centered at 90° azimuth. The BTE placement has a similar decrease in ipsilateral gain for frequencies ≥7 kHz. However, the azimuths of the local minima are not constant across frequency, suggesting destructive interference from head reflections. When the present analysis is expanded to 90°–180° azimuth (not shown), the BTE placement has a monotonic contralateral gain-azimuth function. This implies that the back of the pinna diminishes the effect of the acoustical bright spot, and that the BTE placement's forward-facing non-monotonicities seen in Fig. 1(A) would not be present if the pinnae were reversed.

Without resolving these non-monotonicities, it is likely that the BTE and BTE2 microphone placement will negatively affect the localization of high-frequency sounds for BICI listeners and cause azimuthal mislocalization similar to the acoustic-hearing results of Macaulay et al. (2010). Since ILDs are the main localization cue for BICI listeners, it is possible that the spatial hearing benefits of ATC microphones observed in previous psychophysical studies are simply a result of the BICI listeners having access to larger ILDs. Evidence of this can be seen in Jones et al. (2016), where they found both an increased ILD range and reduced localization error for simulated ATC BICI microphones. Un-synchronized bilateral automatic gain controls are often criticized for their antagonistic relationship with ILDs and BICIs (e.g., Dorman et al., 2014). While previous research clearly supports a role for un-synchronized bilateral automatic gain controls diminishing sound localization performance, the present study reveals that there are other factors affecting sound localization cues like ILDs before any signal processing is applied.

ILDs are more complicated than simply functions of frequency and azimuth. They are also functions of distance and elevation (Wightman and Kistler, 1992), meaning that a single ILD will likely not point to a specific veridical position in space, especially without the aid of ITDs. One approach for improving spatial-hearing benefits to BICI users is to provide binaural cues more similar to that of acoustic-hearing listeners. However, the mechanisms of hearing in BICI listeners are fundamentally different than that of acoustic-hearing listeners (Wightman and Kistler, 1992; van Hoesel and Tyler, 2003; Aronoff et al., 2010). What may be “natural” binaural cues for an acoustic-hearing listener may not be meaningful within the limitations of electric hearing, as the natural binaural cues may not maximize performance. In Brown (2018), real-time corrective processing was applied to BICIs to convert instantaneous ITD to ILD. This processing strategy created ILDs that were larger than those normally observed, and it resulted in a significantly reduced lateralization error for all listeners. This further supports the idea that having a larger range of available ILDs improves performance, not simply more natural ILDs. Additionally, a larger range of ILDs results in larger changes of ILD magnitude between subsequent spatial locations. BICI users may be able to more easily detect changes in spatial location as the changes in ILD magnitude are larger.

Finally, these results have implications for clinical practice. While the 50/50 BTE/ATC microphone blend is similarly monotonic to the ITC and ATC microphone placements (Fig. 2), it has significantly smaller ILDs [Fig. 1(A)]. In order to maximize the ILDs delivered to BICI listeners and to provide more reliable azimuthal localization cues, clinicians should consider programming BICI patients' microphone blends to be 100% ATC microphones, if available.

4. Conclusion

Microphone placement significantly affected the ILDs as would be delivered to BICI listeners. Specifically, ATC and ITC microphones delivered larger ILDs and more monotonic ILD-azimuth functions at high frequencies. These results are a combination of several acoustic factors including the acoustical bright spot, acoustic axis, and microphone ports affecting gain at both the ipsilateral and contralateral processors. ATC microphone placement restores monotonicity to ILD-azimuth functions at high frequencies, and thus should provide a more reliable azimuthal localization cue for BICI listeners.