Introducing Short Interpulse Intervals in High-Rate Pulse Trains Enhances Binaural Timing Sensitivity in Electric Hearing

Abstract

Common envelope-based stimulation strategies for cochlear implants (CIs) use relatively high carrier rates in order to properly encode the speech envelope. For such rates, CI listeners show poor sensitivity to interaural time differences (ITDs), which are important for horizontal-plane sound localization and spatial unmasking of speech. Based on the findings from previous studies, we predicted that ITD sensitivity can be enhanced by including pulses with short interpulse intervals (SIPIs), to a 1000-pulses-per-second (pps) reference pulse train. We measured the sensitivity of eight bilateral CI listeners to ITD while systematically varying both the rate at which SIPIs are introduced (“SIPI rate”) and the time interval between the two pulses forming a SIPI (“SIPI fraction”). Results showed largely enhanced ITD sensitivity relative to the reference condition, with the size of the improvement increasing with decreasing SIPI rate and decreasing SIPI fraction. For the lowest SIPI fraction, insertion of extra pulses brought ITD sensitivity to the level measured for low-rate pulse trains with rates matching the SIPI rates. The results appear promising for the goal of enhancing ITD sensitivity with envelope-based CI strategies by inserting SIPI pulses at strategic times in speech stimuli.

INTRODUCTION

Bilateral cochlear implantation is increasingly used as a clinical treatment for profound hearing loss or deafness. One prominent advantage is that listeners with bilateral cochlear implants (CIs) may be able to use information from both ears to localize sound sources more accurately compared to listeners with unilateral CIs. For sound localization in the left/right dimension, normal-hearing (NH) listeners rely on binaural cues, i.e., interaural time differences (ITD) and interaural level differences (ILD) (for review, see Grothe et al. 2010). The general view is that ITDs provide more salient localization cues at low frequencies and ILDs provide more salient cues at high frequencies (Kistler and Wightman 1992; Macpherson and Middlebrooks 2002; Strutt 1876). Macpherson and Middlebrooks (2002) further showed that for broadband sounds, ITD at low frequencies has higher perceptual weight than ILD. For high-frequency sounds, envelope modulations increase the ITD weight. ITD cues, besides other spatial cues, contribute to spatial release from masking in an informational masking environment (Kidd et al. 2010). Given this, the robust access to ITD information of bilateral CI users would aid sound localization and speech understanding in challenging situations such as multispeaker environments.

Using envelope-based high-rate clinical stimulation strategies such as continuous interleaved sampling (CIS, (Wilson et al. 1991) or advanced combination encoder (ACE, Kiefer et al. 2001), CI listeners have been shown to rely much more on ILDs than ITDs in azimuthal sound localization (Aronoff et al. 2010; Seeber and Fastl 2008; van Hoesel 2004). In principle, CI users could perceive ITD cues encoded via the envelope of modulated stimuli (envelope ITD, ITD_ENV), provided the envelope has a peaked shape (Laback et al. 2004; Laback et al. 2011; Noel and Eddington 2013; van Hoesel et al. 2009). However, speech stimuli processed with envelope-based strategies were shown to provide no salient ITD_ENV cues to CI listeners (Laback et al. 2004; Grantham et al. 2008).1 Another issue is that envelope-based CI strategies do not provide any ITD cues based on the timing of the carrier pulses (see Laback et al. 2015). While CI listeners are sensitive to ITD cues in unmodulated pulse trains when presented via a precisely controlled research system, their sensitivity strongly declines for rates exceeding a few hundred pulses per second (pps) (Laback et al. 2015). Thus, even if stimulation strategies did encode fine structure ITD cues in the carrier pulses, CI listeners would not be sensitive to these cues with the high pulse rates that are required to faithfully transmit speech envelope information (Loizou et al. 2000; Arora et al. 2009).

In order to provide salient ITD cues in a high-rate strategy, Laback and Majdak (2008) proposed to coherently jitter the timing of stimulation pulses presented to the two ears. The original motivation was the finding that NH listeners show poor sensitivity to ongoing ITD (after the signal onset) in high-frequency filtered stimuli with envelope rates exceeding a few hundred hertz (Hafter and Dye 1983) but that introducing a change in the ongoing signal restored ITD_ENV sensitivity (Hafter and Buell 1990). Laback and Majdak (2008) hypothesized a similar effect in electric hearing and compared sensitivity of bilateral CI listeners between high-rate electric pulse trains that were either periodic or involved jitter of each pulse timing (with a rectangular distribution, the width described by a parameter k, ranging from 0 (no jitter) to 1 (maximum jitter)) that was synchronized between the two ears so as to preserve the ITD conveyed by individual pulses. Their listeners showed substantial improvements in ITD sensitivity for the jittered stimuli, particularly at high pulse rates where listeners were not sensitive to ITD with periodic pulse trains. Hancock et al. (2012) replicated the beneficial effect of jitter in single-unit recordings at the inferior colliculus (IC) of bilaterally implanted cats. Jitter restored ongoing firing in most neurons that did not respond to periodic high-rate pulse trains and enhanced neural ITD sensitivity in about half of the IC neurons. The spiking patterns were found to be highly phase-locked, with spikes tending to occur after the occurrence of very short interpulse intervals (IPI). These findings suggest that it is not jitter per se that causes improvements in neural and perceptual ITD sensitivity, but the occasional occurrence of short interpulse intervals (SIPI) in the jittered pulse train. If so, ITD sensitivity may be improved by strategies in which SIPIs are deliberately introduced at selected times.

In the present study, we directly tested the idea that adding extra pulses creating SIPI in a periodic high-rate electric pulse train enhances psychophysically measured ITD sensitivity of human bilateral CI listeners.

In experiment 1, we systematically varied the rate and relative timing of SIPI pulses in order to better understand the mechanism underlying their effect and derive parameter combinations that maximally enhance ITD sensitivity. For comparison, we also tested periodic pulse trains with low rates (referred to as low-rate pulse trains) matching the SIPI rates tested. This allowed us to test the hypothesis that adding SIPI pulses to a high-rate pulse train brings ITD sensitivity to a performance level corresponding to the performance for SIPI pulses presented in isolation. In experiment 2, we aimed to further elucidate the mechanism underlying the SIPI effect by testing how the enhancement of ITD sensitivity is mediated by hypothetical amplitude modulation (AM) cues resulting from changes in short-term electrical power delivered due to SIPI pulse insertion.

EXPERIMENT 1

Listeners and Apparatus

Eight listeners (five females, three males) participated in the experiment. They had been bilaterally implanted with MED-EL devices having 12 electrodes each. The inclusion criteria for participation in the study were good speech perception in daily communication and some basic (better than chance) ITD sensitivity for low-rate pulse trains. Individual listener data are shown in Table 1. One listener fulfilled the inclusion criteria but showed no sensitivity in the jitter condition, even after repeated measurements, and was therefore not tested further. All listeners were paid an hourly wage for their participation. All procedures involving human subjects were approved by the ethics board of the Medical University of Vienna (vote #2155/2013).

TABLE 1.

Participant information regarding age of onset of deafness, implantation of the CI, and electrode pair used for testing. Age of onset of deafness refers to the earlier of the years of onset of deafness in the two ears and may also refer to the age at the approximate onset of profound hearing loss when no specific date of onset of deafness could be defined (e.g., in the case of progressive hearing loss)

Listener	Implants (L and R)	Age at testing(years)	Etiology	Age at onset of deafness (years)	Age at implantation (years)		Duration of bilateral stimulation (years)	Electrode pair
					L	R		L	R
CI1	C40+	31	Meningitis	14	14	14	17	11	10
CI8	C40+	52	Osteogenesis imperfecta	27	41	39	11	10	9
CI11	Concerto (L); pulsar (R)	37	Temporal bone fracture	20	22	22	15	9	9
CI12	C40+	49	Vestibular aqueduct syndrome; progressive	29	35	33	14	10	10
CI24	C40+	53	Progressive	39	41	43	10	8	9
CI60*	Pulsar	68	Meningitis	58	58	58	9	9	9
CI61	Concerto	73	Progressive	Adult	70	69	3	8	7
CI62	C40+	14	Connexin 26	0	2	0	12	11	11
Mean		47.1			35.4	34.8		9.5	9.3

Stimuli were generated on a personal computer and presented to the CIs via a research interface (RIB2) developed at the Institute of Ion Physics and Applied Physics, Leopold-Franzens University of Innsbruck, Austria. The RIB2 allows direct and interaurally coordinated stimulation of two CIs.

Stimuli and Conditions

Biphasic pulse trains of 600-ms duration including linear on- and off-set ramps of 150 ms each were used. Each phase of a pulse had a duration of 26.7 μs by default, but for some listeners, it was increased for certain electrodes to achieve a sufficient maximum (highest) comfortable level (MCL). The MCL is the loudest that the listener can tolerate. The pulse trains were presented at an interaurally pitch-matched electrode pair, which was selected in pretests. A periodic pulse train with a rate of 1000 pps, referred to as high-rate reference, served as the main control condition.

In the SIPI conditions, extra pulses were introduced in the reference pulse train as illustrated in Fig. 1. Each SIPI condition is defined by two parameters: the SIPI rate and the SIPI fraction. The SIPI rate is the rate at which extra pulses are inserted (ranging from 50 to 200 pps). The SIPI fraction is the time interval between the extra pulse and the preceding pulse in the reference pulse train, expressed as a percentage (ranging from 6 to 50 %) of the period of the reference pulse train. As a second control condition, we tested 1000-pps pulse trains with binaurally synchronized timing jitter as used by (Laback and Majdak 2008). We tested a k of 0.9 with the jitter pattern generated freshly for each trial. The third class of control stimuli were low-rate, periodic pulse trains whose rates matched the SIPI rates tested (ranging from 50 to 200 pps). Table 2 summarizes the parameters of the stimuli used in experiment 1.

Fig. 1.

(Color online): Schematic representation of stimuli from experiment 1. Top: high-rate reference pulse train. Bottom: high-rate pulse train with included SIPI pulses

TABLE 2.

Experimental and reference conditions tested in experiment 1. Refer to Fig. 1 and text for explanation of the parameters mentioned. The equivalent SIPI rate for the high-rate jitter was computed as the proportion of all SIPI (≤ 50 % IPI) to all IPI in the jittered pulse train for a jitter factor (k) of 0.9. The SIPI fraction mean and SD were computed considering all SIPI (≤ 50 % IPI) durations

Condition	Rate (pps)	SIPI rate (pps)	SIPI fraction (%)
High-rate reference	1000	0	0
High-rate jitter	1000	222.22 (equivalent)	30 (SD 11.55)
Short IPI	1000	200	6
Short IPI	1000	200	10
Short IPI	1000	200	20
Short IPI	1000	200	50
Short IPI	1000	100	6
Short IPI	1000	100	10
Short IPI	1000	100	20
Short IPI	1000	100	50
Short IPI	1000	50	6
Short IPI	1000	50	10
Short IPI	1000	50	20
Short IPI	1000	50	50
Low rate	200	0	0
Low rate	100	0	0
Low rate	50	0	0

Pretests

The first aim of the pretests was to determine an interaurally place-matched electrode pair for presenting the experimental stimuli. Based on previous findings that pitch-matched electrodes are more likely to elicit better ITD sensitivity than nonmatched electrodes (Kan et al. 2015; Long et al. 2003; Poon et al. 2009; van Hoesel 2004), we used a pitch-matching paradigm. While other methods have been proposed to identify interaural electrode pairs with the best ITD sensitivity (Hu and Dietz 2015), the pitch-matching paradigm was considered sufficient for the goals of our study. This pretest was omitted for listeners who had already performed pitch-matching tests in some of our previous studies. The second aim of the pretests was to determine loudness-balanced stimulation levels for the various experimental stimuli.

Electric Dynamic Range, Pitch Matching, and Loudness Balancing

Unmodulated 300-ms periodic pulse trains at a rate of 1515 pps were used for the first three steps listed below. This high rate was used in order to reduce the potentially confounding effect of rate pitch at lower rates.

In the first step, the electric dynamic range, determined by the threshold (THR) and MCL, as well as a middle level that was judged as a comfortable level (CL) for long periods of time was determined for each electrode in both ears using a graphical loudness scale with a verbal category “middle loud” at the center of the scale. We used a printed scale with verbal attributes for the threshold level, middle level, and too loud level and asked the listener to point on the scale. The CL was used for the subsequent steps in this subsection.

Second, a magnitude estimation procedure was used to estimate the perceived pitch across the electrodes at both ears. Stimuli were presented monaurally in random order at either ear and at each of the available electrodes. Listeners were instructed to assign numbers without any restrictions, according to the perceived pitch of each stimulus. Eight or nine candidate interaural electrode pairs were then chosen for a subsequent pitch-ranking procedure.

Third, listeners compared members of each candidate interaural electrode pair to indicate which electrode elicited a higher pitch percept. A two-interval, two-alternative forced-choice procedure was used. The pitch-matched pair was selected from the basal region of the cochlea by requiring the pitch-ranking score for that pair to be at chance level. For more details on the pitch-matching paradigm, see Majdak et al. (2006).

Fourth, stimuli were presented at both ears simultaneously at 1000 pps to check their binaural loudness while instructing the listeners to attend and respond to the overall loudness by indicating on the visual scale at the perceived loudness, similar to the first step. Starting at 80 % of the monaural CLs, levels were varied simultaneously in proportionally equal steps at the two ears to finally arrive at the most comfortable binaural CL. We checked, using a visual horizontal bar with an indication of the midline, if the auditory image was perceived as centered. If that was not the case, level adjustments were made until the listeners reported a centered auditory image at a comfortable level. These final adjusted CLs were used as the reference in the loudness-matching procedure.

ITD Pretest

The ITD pretest was performed to evaluate whether the listeners were sensitive to ITD cues at the selected pitch-matched electrode pair. The stimulus was a 100-pps pulse train (“Stimuli and Conditions” section), a rate at which listeners usually show the best ITD sensitivity. The results were also used to guide the selection of the range of ITDs to be used in the main ITD experiment. Further, the pretests served as preliminary training for the listeners.

An adaptive two-interval, two-alternative forced-choice procedure was used to measure the threshold for ITD-based left/right discrimination. Visual feedback was presented after each trial. A trial consisted of two intervals separated by a 250-ms silent interval. The first interval contained the reference stimulus with zero ITD. The second interval contained the target stimulus with nonzero ITD, i.e., the pulses at one ear were delayed relative to the other ear. The listeners had to indicate to which side (left or right) the second stimulus was perceived compared with the first stimulus by pressing the corresponding button on a response pad. The ITD of the target stimulus was applied with equal a priori probability, either to the left or to the right side. The procedure started with a target ITD of 800 μs. If a listener showed difficulties with the task, the starting ITD was increased to 1500 μs. The adaptive ITD adjustment followed a three-down one-up rule converging at the 79 % point of the psychometric function (Levitt 1971). The initial step size was 500 μs. After each reversal, it was decreased by a factor of 0.7 until it reached the minimum step size of 50 μs. Each run was finished after 12 reversals, and the ITDs from the last eight reversals were arithmetically averaged yielding the ITD threshold. Some of the ITD thresholds were measured in blocks of two interleaved adaptive runs to compensate for response variance.

Loudness Matching Across Conditions

The perceived loudness of an electric pulse train is known to increase with increasing pulse rate (McKay and McDermott 1998; Shannon 1985). However, the effect of adding SIPI pulses (at various SIPI rates and SIPI fractions) on loudness is not easily predictable from existing loudness models (McKay and McDermott 1998). In order to avoid confounding effects of loudness on our ITD experiments (e.g., Egger et al. 2016, 2017), we performed a formal loudness-matching test using all experimental stimuli, including SIPI and low-rate conditions.

An adaptive loudness-matching procedure was performed to match the loudness of each of the experimental conditions to the reference high-rate condition. A double staircase procedure was used (Jesteadt 1980), consisting of upward and downward staircases. For each experimental condition, the starting levels of the upward and the downward staircases were 20 % below and 20 % above the level of the reference high-rate condition, respectively. The reference and the comparison stimuli were presented in random order in the two intervals of a trial with a silent interval of 200 ms. The listeners reported which of the two stimuli was perceived as louder by pressing the corresponding button. The initial step size was 10 %, and the step size was reduced by a factor of 0.7 after each reversal until it reached the minimum of 2 %. The procedure ended after 12 reversals, and the last six reversals from the upward and the downward staircases were averaged to estimate the loudness-matched level. At least two runs were performed to obtain the final loudness-matched stimulation levels which were used for the different experimental conditions in the main experiment. Because the results of this pretest are potentially interesting for the general understanding of loudness perception in electric hearing, they are analyzed and discussed in the “LOUDNESS EFFECTS” section.

Procedure for Main ITD Experiment

The procedure for the main experiment employed a constant-stimuli paradigm in which a trial consisted of two intervals separated by a 250-ms silent interval. The first interval contained a pulse-train stimulus with zero ITD. The second interval contained the same stimulus with nonzero ITD. The listeners had to indicate to which side (left or right) the second stimulus was perceived compared with the first stimulus by pressing the corresponding button on a response pad. The ITD of the target stimulus was applied with equal a priori probability either to the left or to the right side. Visual feedback on the accuracy of the response was presented after each trial. At least four ITDs were tested per condition. ITDs of 200, 400, and 800 μs were used for all listeners, and additional ITDs (50, 100, and 1600 μs) were selected depending on the listener’s sensitivity determined in the pretests. Sixty trials were presented per ITD per condition, 30 of which with the target stimulus to the left. All trials from all conditions were pooled into a single set, and the order of trials was randomized. The set was then divided into 18 blocks of approximately 250 trials each. Each block lasted approximately 20 min and the listeners took breaks between blocks ad libitum. Testing was done over 2 days.

Results

General

While there are considerable differences in absolute performance across individuals (see Table 3 for listener-specific 75 % ITD thresholds estimated from psychometric functions using a logistic fitting curve2 for selected experimental conditions), the overall pattern of results across different test conditions is consistent across listeners. The participants were not significantly biased in their responses in the left/right discrimination task (Student’s t statistic, t(7) = 0.42, p = 0.41). The amount of variance explained by the logistic curve fit was quantified for each listener and the selected experimental conditions as a percentage, with the average across the listeners and the conditions amounting to 91.47 % (SD 8 %). We therefore focused on the pooled results across listeners and performed statistical analysis on the pooled data. The left panel of Fig. 2 shows the mean scores across listeners (with 95 % confidence intervals) as a function of the ITD, for the different SIPI rates while pooling the data across tested SIPI fractions. The high-rate reference condition and the jitter condition are included for comparison. In general, the performance improved with increasing ITD. Most importantly, for all SIPI rates (empty triangles and diamonds), the performance was clearly better than for the reference condition. Differences between the SIPI conditions and the jitter condition emerged only for ITDs larger than 100 μs. In order to simplify the interpretation of the data, the right panel of Fig. 2 shows performance averaged over the ITDs 200, 400, and 800 μs. The open circles show the mean performance as a function of the SIPI rate, with the results pooled across ITDs and all SIPI fractions. With decreasing SIPI rate (from 200 to 50 pps), the performance improved and approached that observed in the jitter condition. The performance in the low-rate condition follows the same trend as the performance in the SIPI rate conditions, but it appears to be overall better and to increase up to the lowest rate tested.

TABLE 3.

Listener-specific ITD thresholds in select experimental conditions. Responses were fitted using a logistic curve and thresholds were computed at 75 % of the psychometric function

Listener	Jitter (μs)	Low rate, 50 pps (μs)	SIPI rate, 50 pps, fraction 10 % (μs)	Reference (μs)	ENH, 50 pps (μs)
CI1	1156	164	644	ND	–
CI8	444	288	355	ND	700
CI11	246	238	182	ND	152
CI12	153	92	117	ND	97
CI24	1085	1375	2022	ND	338
CI60*	ND	–	ND	ND	–
CI61	323	197	359	ND	88
CI62	165	73	115	ND	–
Mean	510	347	542	ND	275

A * indicates that the listener data was not included in the statistical analyses. A “–” indicates conditions which were not run for that participant due to time constraints

ENH enhanced condition in experiment 2, ND nondeterminable (if performance did not reach 75 %)

Fig. 2.

(Color online): Results of experiment 1; left panel: left/right discrimination performance across different ITDs, for the three SIPI rates (empty symbols connected with solid lines), the high-rate reference condition (filled circles), and the jitter condition (empty circles) condition. Results are pooled over SIPI fractions. Right panel: performance as a function of SIPI rates, pooled over SIPI fractions and ITDs (empty circles). For comparison, performance in corresponding low-rate conditions is shown with solid circles. The dash-dotted and dashed horizontal lines show performance in the jitter and reference conditions, respectively. Error bars around points or lines represent 95 % confidence intervals

Figure 3 shows the effect of SIPI fraction with results pooled over the tested SIPI rates. The left panel shows the results as a function of ITD, with the SIPI fraction as the parameter. Compared to the reference condition, the performance improved with decreasing SIPI fraction at all ITDs. Even for the largest SIPI fraction of 50 % (downward-pointing triangles), the performance appears to be better than for the reference condition. The right panel of Fig. 3 simplifies these data by pooling performance over SIPI rates and ITDs. The improving performance with decreasing SIPI fraction appears to converge at the performance observed in the jitter condition.

Fig. 3.

(Color online): Results of experiment 1; left panel: left/right discrimination performance across different ITDs, for the four SIPI fractions (empty symbols connected with solid lines), the high-rate reference condition (filled circles), and the jitter condition (empty circles). Results are pooled over SIPI rates. Right panel: performance as a function of SIPI fractions, pooled over SIPI rates and ITDs (empty circles). The dash-dotted and dashed horizontal lines show performance in the jitter and reference conditions, respectively. Error bars around points or lines represent 95 % confidence intervals

To examine combinations of the two SIPI parameters and test for a potential interaction between them, Fig. 4 shows the performance as a function of SIPI rate with SIPI fraction as the parameter. The data were pooled over ITDs of 200, 400, and 800 μs. With decreasing SIPI fraction, performance improved at all SIPI rates by about the same amount, saturating at the SIPI fraction of 10 %. Overall, there is no indication of an interaction between the SIPI rate and the SIPI fraction. The best performance was achieved for the combinations of SIPI rates of 50 and 100 pps and SIPI fractions of 6 and 10 %. For comparison, also the performance in the low-rate conditions is shown. At all SIPI rates, the performance in the SIPI fractions 6 and 10 % was not different from the corresponding low-rate performance. The saturation in performance for SIPI rates lower than 100 pps occurs for all SIPI fractions but is not apparent for low-rate pulse trains.

Fig. 4.

(Color online): Left/right discrimination performance as a function of SIPI rate with SIPI fractions as the parameter (empty symbols). Each point represents the performance pooled over ITDs for a certain SIPI rate and SIPI fraction. The performance in the low-rate condition is shown with filled circles. Error bars represent 95 % confidence intervals

Statistical Analysis

The statistical significance of the effects described above was assessed using repeated-measures (RM) analyses of variance (ANOVA) and subsequent post hoc tests (multcompare, MATLAB, Mathworks Inc. using the Bonferroni correction for multiple comparisons). All statistical analyses were performed with a significance criterion of 0.05. Only results for ITDs 200, 400, and 800 μs were included because only these ITDs were available for all participants.

A three-way RM-ANOVA was conducted with the factors SIPI rate, SIPI fraction, and ITD. All three main effects were significant: SIPI rate (F_{2, 207} = 7.79, p < 0.001), SIPI fraction (F_{3, 207} = 26.29, p < 0.001), and ITD (F_{2, 207} = 26.3, p < 0.001). There were no significant interactions (neither two- nor three-way) between the three factors. Post hoc tests showed that performance significantly improved with decreasing SIPI rate from 200 to 100 pps but not further toward 50 pps. Performance significantly improved with decreasing SIPI fraction from 50 to 10 % but not further toward 6 %.

To more closely analyze the overall effect of the SIPI rate, a one-way RM-ANOVA compared the performance for the different SIPI rates with the performances for the high-rate reference and the jitter conditions. The data were pooled across SIPI fractions and ITDs, because these parameters had shown no interaction with the SIPI rate. There was a significant main effect of SIPI rate (F_{4, 274} = 20.59, p < 0.001). Post hoc tests showed that all three SIPI rates had significantly better performance than the reference condition and that only the 200-pps SIPI rate yielded significantly worse performance than the jitter condition.

To more closely inspect the overall effect of the SIPI fraction, another one-way RM-ANOVA compared the performance for the different SIPI fractions with the performances in the high-rate reference and jitter conditions. The data were pooled across SIPI rates and ITD. There was a significant main effect of SIPI fraction (F_{5, 273} = 39.45, p < 0.001). Post hoc tests showed that all four SIPI fractions gave significantly better performance than the reference condition and only the 20 and 50 % SIPI fractions had significantly worse performance than the jitter condition.

Lastly, to compare the low-rate conditions to the SIPI conditions, we added the low-rate conditions as an additional factor level to the factor SIPI fraction and performed a two-way RM-ANOVA on the factors SIPI fraction and SIPI rate, while pooling the data across ITDs. There were significant main effects of both SIPI rate (F_{2, 285} = 17.23, p < 0.001) and SIPI fraction (F_{4, 285} = 34.68, p < 0.001) but no significant interaction. Post hoc tests revealed that only the 50 and 20 % SIPI fractions had significantly lower performance than the low-rate conditions. In other words, for the 6 and 10 % SIPI fractions, the performance for the SIPI conditions does not differ from that in the low-rate conditions.

Discussion

As expected, the performance in the reference condition was essentially at chance level. This is consistent with previous studies showing that CI listeners show little or no ITD sensitivity for unmodulated pulse trains if the pulse rate exceeds a few hundred pulses per second (see recent reviews by Laback et al. (2015) and Kan and Litovsky (2015)). Second, the results for our control condition with timing jitter are consistent with the results of Laback and Majdak (2008), showing large improvements in ITD sensitivity compared to the reference condition. Third, the SIPI conditions also yielded large improvements in ITD sensitivity over the reference condition. The amount of improvement depended on each of the two parameters SIPI rate and SIPI fraction. Specifically, the performance improved with decreasing SIPI rate and decreasing SIPI fraction. No systematic interaction in the effects of the two parameters was found. The best performance was obtained for SIPI rates of 50 and 100 pps combined with SIPI fractions of 6 and 10 %. The lack of an interaction between these parameters provides straightforward rules for applying SIPI pulses to enhance ITD sensitivity in high-rate stimulation strategies. Interestingly, even the least effective SIPI fraction of 50 % or the least effective SIPI rate of 200 pps could significantly improve ITD sensitivity compared to the reference condition.

Comparison of performance with low-rate pulse trains revealed that for SIPI fractions of 6 and 10 %, the performances across SIPI rates (200, 100, and 50 pps) were similar to those for the corresponding low-rate conditions. This indicates that inserting extra pulses with sufficiently short interpulse intervals can produce the same ITD sensitivity as if the extra pulses were presented alone (not embedded in a high-rate pulse train). However, there was only a weak and nonsignificant correlation between listener-specific ITD thresholds in the SIPI and the low-rate conditions (R² = 0.19, p = 0.55), suggesting that the performance in the SIPI conditions may not be predicted by the low-rate performance alone (see Table 3).

The performance for lower SIPI rates (50 and 100 pps) combined with lower SIPI fractions (6 and 10 %) was comparable to performance in the jitter condition (compare Figs. 2 and 4). Additional support for this comes from an analysis of the listener-specific 75 % ITD thresholds (Table 3) showing a highly significant correlation between the jitter condition and those SIPI conditions (R² = 0.92, p < 0.01 for 6 % SIPI and jitter and R² = 0.95, p < 0.01 for 10 % SIPI and jitter). Interestingly, the shortest IPI in the jittered pulse train was 100 μs (with a jitter factor of 0.9 and nominal interpulse interval of 1000 μs, the interpulse interval varies between 100 and 1900 μs), so there was only one SIPI pulse at the SIPI fraction found to be most effective (10 %). This comparison suggests two interpretations: A single SIPI over the course of the jittered pulse train (with a SIPI fraction of 10 %) already triggers sufficient ITD coding and larger SIPIs are not required at all. This implies that a single SIPI pulse per stimulus, in our case yielding a SIPI rate of as low as 3 pps, would be equally efficient as a SIPI rate of 50 pps. This may be tested in future studies. The second interpretation is that a greater number of double-pulse sets with slightly larger SIPIs contributed all together to determine overall ITD sensitivity.

Our finding of greatest performance improvements for short IPI fractions is consistent with the findings of Hancock et al. (2012) who studied the response of IC neurons of bilaterally implanted cats to electric pulse trains with and without timing jitter. While periodic high-rate pulse trains evoked only onset responses and little or no ITD sensitivity, jitter restored ongoing neuronal firing, comparable with that for low-rate pulse trains without jitter, and to enhanced ITD sensitivity in binaurally sensitive IC neurons. More important for the current study, a spike-triggered response averaging analysis they performed suggested that the restoration of ongoing firing was “triggered” by very short IPIs in the jittered pulse trains. Because the preferred spiking times for jittered pulse trains were highly reproducible across stimulus repetitions for an individual neuron and remarkably similar across neurons, they suggested that jitter restores ongoing firing by a relatively low-level, deterministic mechanism. They suggested that nonlinear temporal summation of postsynaptic potentials after a SIPI may drive the neural membrane potentials to cross the adapted firing threshold. In a more recent study (Buechel et al. 2015a), the hypothesis that SIPI pulses can improve neural ITD sensitivity was tested directly in unanesthetized rabbits, using stimuli very similar to ours. Similar to our results, introducing SIPIs in high-rate carriers uncovered latent ITD sensitivity comparable to that observed with low-rate periodic pulse trains. Their results further showed that firing rates tend to increase with decreasing SIPI fraction, in good qualitative agreement with our study.

A different view of the SIPI effect is that the SIPI pulse pair is processed like a single pulse with an enhanced amplitude. In this view, our uniform amplitude pulse train with SIPI is “internally” represented as an AM signal with the modulation rate equal to the SIPI rate. Such AM would likely improve ITD sensitivity, given that CI listeners are sensitive to ITD in low-rate AM high-rate carrier pulse trains (e.g., Laback et al. 2011; van Hoesel et al. 2009). Considering the SIPI pulses as effectively introducing AM cues would be in line with our finding that the ITD sensitivity improvement diminished with increasing SIPI fraction since the modulation depth of the internal stimulus representation of two consecutive pulses is expected to become smaller with increasing distance between the two pulses. In experiment 2, we tested the hypothesis of internal AM representation.

EXPERIMENT 2

In this experiment, we tested the hypothesis that inserting SIPI pulses in a high-rate pulse train enhances ITD sensitivity by means of introducing sharp AM cues in the signal’s “internal” representation. While SIPI pulses may convey enhanced ITD sensitivity by some mechanism depending on their particular temporal pattern, their effect may also be understood by assuming that short-term changes in electric power create AM-like behavior in the neural response via a temporal summation mechanism.

In order to test the potential importance of the temporal structure, we designed SIPI pulse trains involving attenuated SIPI pulse amplitudes such that the short-term power was kept constant across time, as in the high-rate reference condition (see top panel of Fig. 5). If such an attenuated SIPI pulse train provides a similar improvement in ITD sensitivity as a constant amplitude SIPI pulse train, the improvement cannot be attributed to AM cues caused by changes in short-term power.

Fig. 5.

(Color online): Schematic drawing of stimuli from experiment 2. Top: high-rate SIPI pulse train with attenuated SIPI (ATT) pulses. Bottom: high-rate amplitude-enhanced (ENH) pulse train

In order to test whether SIPI pulses introduce AM by short-term power changes, we also tested a condition for which each pair of SIPI pulses was replaced by a single amplitude-enhanced pulse (see bottom panel of Fig. 5). This enhanced condition essentially tested whether the effect of SIPI pulses is comparable to presenting single pulses with larger amplitudes.

Methods

Five listeners (three females, two males) from experiment 1 participated in this experiment.

The stimuli were similar to those described in experiment 1. Only a SIPI rate of 50 pps combined with a SIPI fraction of 10 % was tested, referred to as the SIPI condition. This combination was chosen because it yielded a robust enhancement of ITD sensitivity in experiment 1.

Overall, the tested conditions were (1) the high-rate reference condition from experiment 1; (2) the SIPI condition; (3) the attenuated SIPI condition (ATT), i.e., standard SIPI condition but with the amplitude of the SIPI pulse pairs corresponding to $1/\sqrt{\left(2\right)}$ of that of the other pulses (− 3 dB); and (4) the enhanced condition (ENH), i.e., a regular pulse trains with single pulses enhanced to $\sqrt{\left(2\right)}$ of the other pulses (+ 3 dB). The $\sqrt{\left(2\right)}$ factor equalizes the short-term power between the ENH and SIPI stimuli since power dissipated through a resistor is proportional to the square of the current. One participant was available for a longer period and was tested on two additional ENH conditions with 1 and 2 dB enhancement of the pulse amplitude. By using the loudness balancing paradigm from experiment 1, the amplitudes of each of the test conditions were adapted so that the stimulus’ loudness matched the high-rate reference condition.

The same measures of ITD sensitivity were employed as in experiment 1. All trials from all conditions (high-rate reference, SIPI, ATT, and ENH) were pooled and the order of the trails was randomized. The whole set of trials was then divided into 4 blocks (~ 250 trials each) that were approximately 20 min long. All other procedures were as in experiment 1.

Results

Figure 6 shows the mean results across listeners in experiment 2. The performances for the high-rate reference and the standard SIPI conditions were consistent with performances for these conditions observed in experiment 1. Like the SIPI condition, the ENH condition produced a large enhancement in ITD sensitivity compared to the reference condition, possibly even exceeding the performance for the SIPI condition. In contrast, the ATT condition yielded similar performance to the reference high-rate condition, being essentially at chance level. Listener-specific 75 % ITD thresholds for the ENH condition are shown in Table 3.

Fig. 6.

(Color online): Results of experiment 2; left/right discrimination performance across different ITDs for the ENH (empty circles), the ATT (empty diamonds), the reference (filled circles), and the SIPI (empty triangles) conditions. Error bars around points or lines represent 95 % confidence intervals

A two-way RM-ANOVA with factors condition and ITD showed significant main effects of test condition (F_{3, 44} = 63.8, p < 0.001) and ITD (F_{2, 44} = 4.36, p = 0.019), but no significant interaction (F_{6, 44} = 1.84, p = 0.113). Post hoc comparisons (multcompare) showed significantly higher performance for both the SIPI and the ENH conditions compared to the reference high-rate condition, but no difference between the SIPI and the ENH condition. The ATT condition was not significantly different from the reference high-rate condition.

The results for the listener who was available for testing additional enhancement conditions showed improvements in performance for the 1-dB ENH condition compared to the reference condition which further increased for the 2-dB condition but not further for the 3-dB condition. The performance for the latter two conditions was comparable to the corresponding SIPI condition.

Discussion

The results show that regularly enhancing the pulse amplitude (ENH condition) produces a similar improvement in ITD sensitivity as adding SIPI pulses at regular intervals (SIPI condition). In both conditions, the observed ITD sensitivity was significantly higher than that in the reference condition. While the performance improvement in the ENH condition seemed to be slightly higher than that in the SIPI condition, the difference between these two conditions was not statistically significant. The similarity of the mean performance across listeners for the ENH and the SIPI conditions appears to be consistent with the hypothesis that both of these stimuli elicited AM in their internal representations. Further evidence for this interpretation comes from the ATT condition, which did not produce any improvement in ITD sensitivity compared to the reference condition. In that condition, the amplitudes of SIPI pulses were attenuated such that the short-term power was kept constant as for the high-rate reference pulse train. These results are consistent with the assumption that the internal (neural) presentation of that condition is similar to that of the reference condition. Overall, our results are consistent with physiological measurements in rabbits using both SIPI and ENH pulse trains (Buechel et al. 2016) and with the conclusions derived by Hancock et al. (2012). Hancock et al. (2012) proposed that the dynamics of low-voltage-activated K⁺ currents (I_{K, LVA}) may play a role in generating responses for very short IPIs in jittered pulse trains. At high rates, these currents raise the spiking threshold, thereby decreasing the firing rate and masking ITD sensitivity. The introduction of either a SIPI or a transient increase in pulse amplitude would then raise membrane potentials sufficiently rapidly to exceed the adapted threshold and elicit spikes.

While the results of the current study suggest that SIPI pulse trains and ENH pulse trains are similarly efficient in conveying salient ITD cues, there are potential differences between the two approaches. First, the finding that SIPI pulse trains with a fraction up to a least 20 % produced significant improvements in ITD sensitivity, even though to a smaller degree, suggests a gradual transition between purely temporal coding (inserting extra pulses with constant amplitude) and AM coding. Such differences may account for the low correlation between listener-specific ITD thresholds for the 50-pps-SIPI and the ENH conditions (Table 3, R² = 0.52, p = 0.48). Second, SIPI pulse trains eliciting robust enhancements in ITD sensitivity produced no increase in loudness, whereas ENH pulse trains produced a significant increase in loudness. In terms of ITD coding in CI systems, this could be an advantage for the SIPI approach, where it is desired to obtain performance improvements at a constant loudness. Third, the SIPI and ENH approaches may differ in their impact on temporal coding of speech, a research topic for future studies. For example, SIPI pulse trains with intermediate SIPI fractions may be comparably efficient in improving ITD sensitivity but less detrimental for speech coding compared to ENH pulse trains. Fourth, the ENH approach is more flexible with respect to pulse timing considerations, particularly when considering across-channel interactions.

LOUDNESS EFFECTS

The primary purpose of the loudness matching of experimental stimuli performed in the pretests of experiments 1 and 2 was to ensure constant loudness across stimulus conditions, which is relevant for clinical application in CI stimulation strategies. Because the effects of the stimulus manipulations on loudness are potentially interesting for a general understanding of loudness in electrical hearing, they are reported and discussed in more detail here. Figure 7 shows the mean loudness-matched amplitudes across listeners (± 95 % confidence intervals) for the different experimental conditions from the two experiments relative to the amplitudes of the high-rate reference condition. Positive amplitudes along the ordinate indicate that larger amplitudes were required in order to match the loudness of the reference condition, meaning that the stimulus with equal amplitude was less loud than the reference.

Fig. 7.

(Color online): Loudness-matched amplitudes of experimental conditions relative to the amplitude of the high-rate reference condition. Amplitudes are plotted separately for the left (empty symbols) and the right (filled symbols) ear. Different symbols indicate different classes of stimuli: jitter (left facing triangles), SIPI (diamonds), ENH (downward facing triangles), and low rates (upward facing triangles). Error bars represent 95 % confidence intervals

An RM-ANOVA comparing the low-rate conditions with the reference condition showed a significant main effect (F_{18, 97} = 10.56, p < 0.001). Post hoc comparisons (multcompare) indicated that both the 50-pps (p < 0.001) and 100-pps (p = 0.0013) conditions required significantly higher amplitudes to match the loudness of the reference condition. A separate RM-ANOVA, comparing all high-rate conditions (jitter, SIPI, ATT, and ENH conditions) to the reference condition, showed a significant main effect (F_{15, 82} = 9.9, p < 0.001). Post hoc comparisons showed significantly lower matched amplitudes for the SIPI fractions of 6 and 10 % at a SIPI rate of 200 pps and for the ENH condition (p < 0.001 for all) compared to the reference condition. For all other conditions, including the vast majority of SIPI conditions, the jitter condition, and the ATT condition, there was no significant deviation in loudness from the reference condition.

The reduction in loudness for the low pulse rates compared to the reference condition is consistent with earlier findings that loudness increases with increasing pulse rate (Carlyon et al. 2015; McKay and McDermott 1998; Shannon 1985). The lack of an effect of binaurally coherent jitter on loudness is consistent with Laback and Majdak (2008), who reported no difference in loudness between periodic and jittered pulse trains at various rates observed from informal loudness scaling. SIPI conditions with low SIPI rates (50 and 100 pps), which were found to be as effective in enhancing ITD sensitivity as jitter, also showed no difference in loudness compared to the reference condition. As discussed above, both jitter and SIPIs have been shown to restore ongoing neural spiking and improve ITD coding in a majority of inferior colliculus neurons (Hancock et al. 2012; Buechel et al. 2015b). Thus, the lack of an effect of jitter and SIPIs on loudness seems inconsistent with the view that loudness is related to the total neural spiking activity elicited by a stimulus (Fletcher and Munson 1933; Lachs et al. 1984). This paradox might be resolved if loudness encoding were dependent on a different set of neurons than those coding ITD. A caveat is that the sampling of neurons in Fletcher and Munson (1933) and Lachs et al. (1984) might have been biased toward larger neurons or toward those responsive to a single electric pulse, and thus might not be representative of total spiking activity. In any case, while already previous studies demonstrated the difficulty in predicting loudness based on physiologically measured spike counts (see Delgutte (1996) and Relkin and Doucet (1997)), a more recent study found that the behavior of nonprimary-like (especially chopper) cochlear-nucleus neurons is predictive of loudness recruitment in noise-exposed animals (Cai et al. 2009).

The equal loudness observed for the ATT, jitter, and the reference conditions, all having the same total power, is consistent with the idea that long-term power is important for loudness. Also, the greater loudness for some of the 200-pps SIPI conditions compared to the SIPI conditions with lower SIPI rates and the greater loudness for the ENH pulse train compared to the reference condition are consistent with this idea since these conditions lead to an increase in long-term power. On the other hand, the 200-pps SIPI rate conditions with larger SIPI fractions (20 and 50 %) were perceived as equally loud as the reference condition, despite their higher power. Taken together, these comparisons clearly suggest that in addition to long-term stimulus power, the temporal structure of stimulation pulses, i.e., temporal variations in short-term power, is important for loudness. The current data may be useful for further evaluating and advancing loudness models of electric stimulation (McKay et al. 2003) in any case.

SUMMARY AND CONCLUSIONS

The experiments described in this study investigated the effects of inserting SIPI pulses into high-rate pulse trains on ITD sensitivity of bilateral CI listeners. Based on previous studies, it was hypothesized that introducing SIPI pulses restores ITD sensitivity at high rates to a performance level achieved with periodic low-rate pulse trains. The rate of introduction of SIPI pulses (SIPI rate) and the interval between inserted pulses and preceding pulses (SIPI fraction) were systematically varied, and the ITD sensitivity was tested in a left/right discrimination task. In general, introducing SIPI pulses enhanced ITD sensitivity relative to the periodic reference condition. Improvements in ITD sensitivity tended to increase with decreasing SIPI rate and with decreasing SIPI fraction. The combination of low SIPI rates (50 and 100 pps) and small SIPI fractions (6 and 10 %) yielded the greatest improvements. Introducing SIPI pulses with small SIPI fractions increased performance to the level found for periodic low-rate pulse trains with rates matching the SIPI rates, thus confirming the hypothesis that in the optimal case SIPI pulses completely overcome pulse rate limitations in ITD perception.

ITD sensitivity was additionally measured for high-rate pulse trains with binaurally coherent jitter. With the optimal SIPI parameters mentioned above, the improvements observed were very similar to those obtained for jittered pulse trains. It appears that the idea of enhancing ITD sensitivity in high-rate CI stimulation strategies with timing jitter (Laback and Majdak 2008) can be replaced with the more controlled and flexible scheme of introducing SIPI pulses at strategic times of the stimulus envelope (Hancock et al. 2012). Studies are underway to test the effects of inserting SIPI pulses into amplitude-modulated high-rate pulse trains.

Results from experiment 2 support the idea that the improvements in ITD sensitivity achieved by inserting SIPI pulses are due to short-term changes in stimulus power, which can be interpreted as introducing AM cues in the stimulus’ neural representation. A 3-dB enhancement in the amplitude of single pulses at a fixed rate (50 Hz), a condition referred to as ENH, produced the same improvements in ITD sensitivity as inserting SIPI pulses at the same rate. Nevertheless, the SIPI approach may still be advantageous for signal coding, with the goal of minimal degradation of speech information. Our loudness-matching data showed that, in contrast to introducing ENH pulses, introducing SIPI pulses would require no reduction in pulse amplitude to maintain constant loudness, which may simplify loudness considerations. The extent to which speech perception would be affected by the introduction of SIPI pulses or by using enhanced pulse trains remains a topic for future research. The goal of such strategies would be to find the optimal trade-off between maximizing ITD sensitivity while minimally degrading monaural speech perception. The two parameters SIPI rate and SIPI fraction allow room for finding such an optimal trade-off. In order to minimize degradation of speech information, SIPI pulses could be introduced at fractions or multiples of the fundamental period of voiced speech.

Because real-life stimuli involve amplitude modulation potentially creating envelop ITD cues, an important question is how SIPI pulses should be inserted in relation to the temporal envelope of the corresponding CI channel in order to provide maximal benefit for ITD perception. Intuitively, introducing SIPI pulses at selected peaks of the envelope appears most promising when simultaneously aiming at preserving the original speech envelope. Studies currently underway appear to support this idea.

With respect to implementing the SIPI approach to multichannel CI stimulation strategies that activate adjacent electrodes in a staggered manner, additional issues need to be considered. For example, stimulation of adjacent electrodes at very short IPIs may trigger highly pulse-locked neural spikes as an unintended effect of channel interaction. Furthermore, an alternative approach proposed to convey salient ITD cues with bilateral CI stimulation strategies should be considered: several studies proposed to encode salient ITD cues by means of using low-rate carrier pulses, at least at the most apical electrodes (e.g., Churchill et al. 2014; FS4: Riss et al. 2014; PDT: van Hoesel and Tyler 2003; FAST: Smith 2010). A promising approach to provide maximal speech and ITD information may be to combine low-rate stimulation at most apical electrodes (where SIPI pulses are likely not beneficial) with strategic insertion of SIPI pulses into high-rate pulse trains presented at the other electrodes. In conclusion, the current results suggest that it is possible to convey salient ITD cues to bilateral CI listeners at high pulse rates that are required for speech coding. Before applying this approach in CI stimulation strategies, several issues need to be addressed, including the optimal timing of SIPIs relative to the speech envelope, the effect of channel interactions, and the impact of SIPI pulses on temporal pitch perception and speech understanding.

Funding Information

We thank the Austrian Academy of Sciences for long-term funding research on spatial hearing with CIs. This study was funded by the National Institutes of Health grant R01 DC 005775. Additional funding was provided by the European Commission (Project ALT, Grant 691229).

Compliance with Ethical Standards

Conflict of Interest

The authors declare that they have no conflict of interest.