Frequency series

4.3.1.1 Control frequencies: 8 – 12.5 kHz

Figures 31a and b show the voltage drive from channel 1 of audiometer 1 to the left HDA 200 headphone for the test frequencies 8, 9, 10, 11.2 and 12.5 kHz, at which the listener did not report hearing any spurious noises or off-test-frequency tones. At each frequency the signal was presented at the threshold of the listener to determine whether any unwanted noise that would have been audible with this particular audiometric configuration were present.

Audiometric thresholds are plotted against the audiometer output for comparison. Figure 31a) shows frequency plotted linearly on the x-axis to improve visibility of higher frequency components of the output and Figure 31b) shows frequency plotted on a logarithmic scale to emphasise low-frequency components in the spectrum of the output. Visual inspection of the data shows that at these relatively low intensity levels, unintended spectral peaks are present at approximately 5700, 10000, 14000, and 18000 Hz, the most prominent of which is centred at 14000 Hz. The peaks are also present in the noise floor recording and therefore seem to be intrinsic in the measurement system. Such tones are less likely to result in false threshold measurements as they are constant throughout the testing, rather than appearing only when the audiologist presents the test tone. However, examination of lower frequency output at the EHF frequencies in Figure 31b, shows an additional three peaks in the power spectrum at 50, 100, and 150 Hz that do not shift in frequency or level depending on the frequency of the test tone, and which are not present in the noise floor recording. Audiometric thresholds were not measured at these frequencies, therefore it is not clear whether this low-frequency noise present only on presentation of the tone could be audible to the listener.

-140 frequencies from 8 to 12.5 kHz. Test tones were presented at the level of the listener’s pure-tone threshold at each frequency. The listener did not report being aware of any noise or off-test-frequency tones during testing of these frequencies.

a) linear frequency axis

b) log frequency axis

4.3.1.2 Problematic frequencies: 14 and 16 kHz 4.3.1.2.1 Audiometer 1

The spectra of the electrical output of audiometer 1 at the level of the listener’s pure-tone thresholds at 14 and 16 kHz are shown in Figures 32 a and b, using the same format described above for Figure 31. With this audiometer, the listener reported hearing audible lower-frequency tones when the 16 kHz test tone was presented, but not when 14 kHz was tested. However, the recordings made at both these frequencies show unwanted noise at frequencies lower than the test tone. Of particular note in the voltage drive spectrum for 14 kHz is the narrow peak at approximately 5880 Hz that rises above the listener’s audiometric thresholds. Although the listener did not report hearing this particular pure-tone-like component when 14 kHz was presented, the recordings suggest it may be audible to listeners with good hearing acuity at this frequency. Also present in the 14 kHz output were peaks at approximately 50, 150, and 18000 Hz similar to those noted in the spectra of the five lower EHF frequencies. None of these additional components of the spectrum were present in the noise floor.

The test-tone presented at 24 dB HL at 16 kHz produced a voltage drive to the headphone that differed markedly from the output recorded at lower frequency test tones. Figure 32a shows prominent spectral peaks associated with stimulus presentation that were recorded at approximately 8000 and 14000 Hz. In particular, the broad peak at around 8000 Hz reaches a level close to the listener’s threshold at that frequency. In Figure 32b, the 50 Hz peak found in the 14 kHz test tone output is again present, as is a peak at around 140 Hz. Also notable is the rise in the output level of the spectrum below around 2000 Hz. The voltage drive gradually increases down to approximately 300 Hz, after which it plateaus at this elevated, potentially audible, level. It was unable to be ascertained which of these lower-frequency tones were heard by the listener, but the fact that they appear only when the test tone is being presented by the audiologist increases the likelihood that false thresholds could be recorded for this frequency.

-140

4.3.1.2.2 Audiometer 2

The measurements at 14 and 16 kHz are shown for audiometer 2 in Figure 33 a and b. Similar to the power spectra of audiometer 1, the electrical output of audiometer 2 for a 14 kHz test tone showed unwanted spectral peaks at around 5900 to 6100 Hz.

The voltage drive from audiometer 2 for a 16 kHz tone has more peaks than that recorded for the same signal from audiometer 1, in particular, the prominent peak at around 8000 Hz.

Above the test frequency, small to medium peaks unlikely to influence the audiogram were present at approximately 17000, 21500 Hz, and 23500 Hz. Below the test frequency, a large peak was recorded at 30 Hz and a very broad increase in energy was present from approximately 700 to 1700 Hz. While the broad increase in output below 2 kHz is present in the voltage drive recordings of audiometer 1 is absent in audiometer 2, the peaks in the low-frequency response of both audiometers for the 16 kHz tone are a concern. At 16 kHz,

“noise” was audible to listeners with both audiometers, however differences in the subjective properties of the “noise” were not assessed.

-140 -120 -100 -80 -60 -40 -20 0

25 250 2500 25000

Voltage drive to HDA 200 headphones (dBV)

Frequency (Hz) Measurement noise floor

14 kHz 30 dB HL 16 kHz 24 dB HL Listener's threshold -140

-120 -100 -80 -60 -40 -20 0

0 5000 10000 15000 20000 25000

Voltage drive to HDA 200 headphones (dBV)

Frequency (Hz) Measurement noise floor

14 kHz 30 dB HL 16 kHz 24 dB HL Listener's threshold

Figure 33. The voltage drive of audiometer 2 to the HDA 200 headphones at 14 and 16 kHz. Test tones were presented at the level of the listener’s pure-tone threshold at each frequency. Note the potentially audible increase in noise around 1200 Hz when the 16 kHz tone is presented.

a) linear frequency axis

b) log frequency axis

4.3.2 16 kHz intensity series 4.3.2.1 Audiometer 1

Figures 34a and b show the voltage drive to the HDA 200 headphones from audiometer 1 for output levels from -10 to the maximum presentation level of 55 dB HL. Stimuli were presented in 10 dB HL steps from -10 to 40 dB HL and in 5 dB HL steps from 45 dB HL to 55 dB HL. As in the frequency series, the output is shown in two panels; with frequency plotted linearly in panel a) and logarithmically in panel b). From these figures it is clear that overall, when the intensity of the test tone was increased in 5 or 10 dB steps, the level of the output across the entire frequency range increased approximately linearly. The exception is the peaks at 50, 100, and 150 Hz, which remain at the same level and frequency until they are covered by the rise in the overall level of the low-frequency output at 40 dB HL for 50 Hz, and at 30 dB HL for 100 Hz and 150 Hz. The spectral peaks at 130 and 8000 Hz are recordable from the lowest intensity level and increase with the rest of the spectrum, so that the level of these peaks above the level of other noise is maintained across all intensity levels.

It is clear from Figure 34 a and b that the frequency noise floor greatly exceeds the low-frequency hearing thresholds of the participant acting as the listener, even at low to moderate presentation levels of the 16 kHz tone.

4.3.2.2 Audiometer 2

In contrast to audiometer 1, when the intensity of the 16 kHz test tone was increased using audiometer 2, the frequency of unwanted spectral peaks shifted. As for audiometer 1, Figure 35a and b show the voltage drive for the audiometer as the intensity of the 16 kHz tone was increased in 10 dB steps from -10 dB HL to 40 dB HL, and in 5 dB steps from 45 dB HL to the maximum output level of 60 dB HL. Overall, the power spectrum of the output is less consistent across intensity than that for audiometer 1. In particular, the centre frequency of peaks at around 400 – 1100, 85000, 165000 and 23000 Hz shift slightly with intensity (Figure 35a). The shift in the centre frequency of peaks is clearly illustrated in Figure 35b, where a large spectral peak shifts from approximately 150 Hz at the lowest stimulus intensity to 100 Hz at the highest intensity. The broad rise in energy around 500 – 1000 Hz also varies markedly with intensity in regards to the centre frequency of the peak. This rise below 2 kHz plateaus at the higher intensities more than at lower intensities, for example at 30 dB HL, the voltage drive initially rises below 2 kHz, but returns to the level of the noise floor by 500 Hz.

The pattern of this increase in energy at higher intensities, particularly 60 dB HL is more similar to the pattern recorded at all intensities for audiometer 1, with no return of the voltage drive to the level of the noise floor with decreasing frequency. Again, the low-frequency noise-floor increases well above the noise floor of the measurement equipment and the pure-tone thresholds of the listener at stimulus presentation levels of 30 dB HL and above.

4.3.3 Repeatability

Repeated measurements of the voltage drive to the HDA 200 headphones at each intensity level at 16 kHz (data not shown) showed no evidence that the unwanted peaks in the output spectrum were intermittent in either audiometer.

-120 noise floor with increasing presentation level of the 16 kHz tone. The listener’s thresholds are plotted for comparison.

a) linear frequency axis

b) log frequency axis

-120 frequency with increasing presentation level of the 16 kHz tone. The listener’s thresholds are plotted for comparison.

a) linear frequency axis

b) log frequency axis

4.4 Discussion

The measurements of the spectra of the voltage drive to the headphones performed for two calibrated GSI 61 audiometers showed clear evidence of unwanted noise in the output when tones of 14 and 16 kHz were presented. The purpose of these measurements was not to conduct an exhaustive investigation of the characteristics of the output of the audiometer at each test frequency and presentation level, but instead to identify whether the spurious noise in the output of the audiometer previously reported by Schmuziger et al. (2007) and Kurakata et al. (2010) was an issue for the particular audiometers being used for EHF testing throughout the studies reported in this thesis. It was considered especially important that we perform these measurements given that one of the investigators, as well as several listeners tested during data collection for the study presented in Chapter 3, reported hearing “other noises” during pure-tone threshold seeking at 16 kHz. Although Schmuziger et al. provided similar subjective reports; they could not identify any unwanted noise in the acoustic output of their GSI 61 audiometer, prompting the need to clarify this issue.

The frequency series conducted showed that when pure-tones at each of the EHF test frequencies were presented at the level of the listener’s thresholds, unwanted noise in the conventional frequency range was primarily a concern at 14 and 16 kHz for both audiometers. Although some spectral peaks at 150 Hz and below were recorded for lower frequency test tones, their impact is unclear as audiometric thresholds were not available for these frequencies. Given our results showing the presence of unwanted sounds with two GSI 61 audiometers and the subjective presence of such noise in Schmuziger et al. (2007) study, it seems likely that the absence of unwanted noise in the acoustic output measurements of their audiometer was a result of limitations in the measurement equipment.

When a 14 kHz pure-tone was presented the feature of primary concern in the output of both our audiometers was a spectral peak at around 5.9 kHz. This peak exceeded the listener’s audiometric threshold, however it could not be detected as a separate tone during testing. For the 16 kHz test tone, concerning characteristics of the spectral output included an increase in both narrow- and broad-band components below approximately 2 kHz. While both audiometers showed unwanted spectral energy below 2 kHz, the frequency and intensity characteristics of this noise varied significantly between the two audiometers, particularly at lower presentation levels. Specifically, the centre frequency of peaks varied markedly across presentation levels in the output of audiometer 2, but not audiometer 1. Additionally, the rise in energy below approximately 2 kHz plateaued to a consistent level across frequencies

below approximately 300 Hz in the output of audiometer 1, but returned to the level of the noise floor at low stimulus intensities for audiometer 2. Overall, the output of audiometer 2 was less consistent and contained more unwanted peaks than the output of audiometer 1.

In agreement with the present results, Schmuziger et al. (2007) showed a rise in the acoustic output of the Madsen Itera II audiometer below about 2 kHz, although they did not comment on this. However, the centre frequencies of other unintended spectral peaks documented by Schmuziger et al. varied from the frequencies of peaks found in the output of either of our audiometers. Also conflicting with the present results, Kurakata et al. (2010) documented spectral peaks at 2 and 4 kHz in the electrical output of their Siemens UNITY 2 audiometer which were not present in our recordings. Given our data showing that the characteristics of unwanted noise vary between two audiometers of the same model, it is not surprising that the properties of the noise would be different in audiometers of different models.

Certainly the presence of unwanted noise in the output of calibrated audiometers is a concern when assessing the validity of EHF measurements; however whether such noise will be a problem during a given test session will depend on the particular audiometric configuration of the listener. Our listener’s audiogram provides a particularly good example of the type of audiogram in which noise is likely to be an issue; thresholds are very good across most frequencies, with a rise to above 20 dB HL only at 14 and 16 kHz. It is this much better sensitivity for unwanted tones than for the test-tone that creates a problem as the lower-frequency tones could be detected before the test tone, leading to the recording of an artificially lower threshold for the test tone. Such an audiometric configuration would be expected to be relatively common given that thresholds typically decrease at the highest audible frequencies first with increasing age (e.g. Ahmed et al., 2001; Stelmachowicz, Beauchaine, Kalberer, Kelly, et al., 1989). Schmuziger et al. (2007) suggest that unwanted noise is more likely to be a concern for listeners with EHF hearing loss than those with normal hearing, as low-frequency tones will usually occur only near maximum output levels of the audiometer. However, our measurements show that unwanted noise is present at relatively low intensity levels, and therefore may be detected when thresholds at the highest test frequencies are similar or only slightly elevated compared to those at lower frequencies.

The differences in the output of the two audiometers of the same model shown in our measurements are particularly concerning. If more than one audiometer is being used to collect data to be compared over time, for example in patients being monitored during

cause threshold changes to be identified that do not reflect true changes in hearing acuity.

This possibility should be noted by any clinician performing monitoring of EHF thresholds. It is somewhat reassuring that we found no evidence that the noise present was intermittent, suggesting that when repeated audiograms are measured using the same audiometer, as they were for all participants tested in Chapter 3, they can be reliably compared over time.

Schmuziger et al. (2007) suggested that the introduction of steady, low-level noise to the test ear could be used to mask unwanted noise or tones and reduce their effect on thresholds.

While this is a good suggestion, it is complicated by the fact that the characteristics of the spurious noise and tones varies across audiometers, and in some cases across presentation levels, necessitating the creation of either individualised masking noise for each audiometer, or the use of a relatively broadband masking noise. If a broadband noise is used, the intensity will need to be higher to effectively mask peaks in the spectrum of the audiometer output, which could potentially increase the difficulty of the task and discomfort for the listener.

Further research is needed to investigate the viability of masking noise as a solution for unwanted peaks in the output of audiometers.

Alternatively, in accordance with IEC 60645-1 (2012), the output of the audiometer at 14 and 16 kHz could be limited to a level at which any noise present will not be audible. Given that noise above normal thresholds was found for a 14 kHz tone presented at 30 dB HL, this method would severely limit our ability to measure any threshold outside the normal range at these high-frequencies and therefore detect hearing loss.

4.4.1 Conclusion

At a minimum, the present results confirm that the issue of unwanted lower-frequency energy in the output of audiometers at 14 and 16 kHz is widespread and should be acknowledged when interpreting EHF audiograms and changes in thresholds at these frequencies over time.

Given that off-frequency tones are not intermittent, we conclude that, when the same audiometer is used for repeated testing, changes in recorded thresholds are likely to be genuine, and not the result of differences in the spectra of the output over time. However, whether changes in 14 and 16 kHz thresholds, particularly in listeners with near-normal conventional frequency thresholds, reflect a true change in hearing sensitivity at 14 and 16 kHz, is less certain.

In the majority of cases of postoperative EHF hearing loss documented in the previous chapter, thresholds at lower frequencies improved. If listeners were responding to newly audible lower frequency tones, we would expect an improvement in EHF thresholds that corresponded with the improvement at conventional frequency thresholds. Given that the opposite of this prediction was true, we conclude that, at least in the majority of patients, EHF threshold changes reported in Chapter 3 were not the result of unintended tones in the audiometer output. However, in cases in which conventional frequency thresholds deteriorated so that unwanted tones that may have influenced the preoperative EHF audiogram became inaudible, we cannot exclude the possibility that inaccuracies in the audiometer output influenced the results. Fortunately, such cases were few in number, particularly following stapedectomy, where the greatest EHF losses were recorded.

Chapter 5: Literature review – Potential causes of extended high-frequency