Department of Physiology, Centre for the Neural Basis of Hearing, Cambridge CB2 3EG, United Kingdom
![]() |
ABSTRACT |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Wiegrebe, Lutz and Ian M. Winter. Temporal Representation of Iterated Rippled Noise as a Function of Delay and Sound Level in the Ventral Cochlear Nucleus. J. Neurophysiol. 85: 1206-1219, 2001. The discharge patterns of single units in the ventral cochlear nucleus (VCN) of anesthetized guinea pigs were examined in response to iterated rippled noise (IRN) as a function of the IRN delay (which determines the IRN pitch) and the IRN sound level. Delays were varied over five octaves in half-octave steps, and sound levels were varied over a 30- or 50-dB range in steps of 5 dB. Neural responses were analyzed in terms of first-order and all-order inter-spike intervals (ISIs). The IRN quasi-periodicity was preserved in the all-order ISIs for most units independent of unit type or best frequency (BF). A deterioration of the temporal all-order code was found, however, when the neural response was influenced by inhibition. The IRN quasi-periodicity was also preserved in first-order ISIs for a limited range of IRN delays and levels. Sustained Chopper units (CS) in the VCN responded with very regular ISIs when the IRN delay corresponded to the unit's chopping period; i.e., the unit showed an increased proportion of intervals corresponding to the IRN delay (interval enhancement) relative to an equal-level, white-noise stimulation. This interval enhancement has a band-pass characteristic with a peak corresponding to the chopping period. Moreover, for CS units in rate saturation, the chopping period, and thus the interval enhancement to the IRN, did not vary with level. Units classified as onset-chopper also show a band-pass interval enhancement to the IRN stimuli; however, they show more level-dependent changes than CS units. Primary-like (PL) units also show level-dependent changes in their ability to code the IRN pitch in first-order intervals. The range of delays where PL units showed interval enhancement was broader and extended to shorter delays. Based on these findings, it is suggested that CS units may play an important role in pitch processing in that they transform a higher-order interval code into a first-order interval place code. Their limited dynamic range together with the preservation of the temporal stimulus features in saturation may serve as a physiological basis for the perceived level independence of pitch.
![]() |
INTRODUCTION |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Psychophysically, the pitch
perception of complex sounds is relatively independent of presentation
level. When the level of a complex harmonic sound is varied from 50 to
80 dB SL, its pitch changes by, at most, 1% of the fundamental
frequency (Zwicker and Fastl 1990). This perceptual
level independence presents a challenge for physiological studies
looking for temporal or spectral correlates of pitch in the mammalian
auditory system: spectral models of pitch perception rely on the
spectral magnitude information as it is preserved in a rate-place code
along the tonotopic neural maps of the auditory system. Temporal models
rely on the preservation of the temporal stimulus characteristics in
the timing of spikes. Both these approaches have been investigated as
to the extent to which they are vulnerable to changes in level (see
Langner 1992
for a review).
Spectral encoding in the periphery
The majority of auditory-nerve fibers have a dynamic range of
about 40 dB, above which the fiber response saturates (Evans and
Palmer 1980; Sachs and Abbas 1974
; Winter
and Palmer 1990
). Across a tonotopic map, saturation decreases
the contrast in the harmonic spectrum due to "clipping" of the
spectral peaks. Sachs and Young (1979)
investigated the
spectral representation of vowel formants across a large number of
auditory-nerve fibers with BFs covering the whole vowel spectrum. They
found that, while fibers with high spontaneous rate were in saturation,
fibers with spontaneous rates below 20 spikes per second preserved the
spectral vowel envelope. They did not investigate, however, the extent
to which the harmonic spectral ripple itself is preserved in a
rate-place code; i.e., they investigated a neural correlate of the
vowel type but not of its fundamental frequency. In further studies, Miller and Sachs (1984)
investigated the encoding of the
fundamental frequency of speechlike sounds, and they demonstrated a
robust code in a temporal-place representation; however, they did not investigate a rate-place code. It is therefore possible that a rate-place code may exist for the fundamental frequency, particularly if one takes into account the responses of high-threshold, low spontaneous rate auditory-nerve fibers. Recent models based on recordings from the cochlear nerve of the cat and transferred to a
human cochlear place map indicate that as many as the first five
harmonics are represented in a rate-place code even at reasonably high
stimulus levels (Delgutte 1996
). Thus it is possible for the fundamental frequency to be signaled to the brain in either a
rate-place or temporal-place code, and therefore the responses of cells
in the cochlear nucleus (an obligatory synapse for the auditory nerve)
are of considerable importance. Some neurons in the cochlear nucleus
are characterized by a dynamic range much greater than individual
auditory nerve fibers; the range can be as much as 80-90 dB. These
neurons were classified as onset-choppers (OC) (Rhode and Smith
1986
; Winter and Palmer 1995
). However, OC units
have a mean bandwidth of typically 3 octaves and may even be as wide as
6 octaves (Jiang et al. 1996
; Palmer et al. 1996
). Thus, although these neurons may be able to represent
the level range within which pitch perception is rather stable without saturation, OC neurons appear not to provide sufficient frequency selectivity to represent the harmonic spectrum.
Temporal encoding in the periphery
The temporal coding of the pitch of steady-state vowel sounds, in
terms of the synchronization index, appears robust as a function of
stimulus level (Miller and Sachs 1984), but it is unlikely that such a code underlies pitch perception as it cannot explain the phenomenon of pitch shift (Schouten 1940
).
Instead, it has been suggested that inter-spike intervals (ISIs) may
provide a better code for the representation of the pitch of complex
tones (e.g., Evans 1978
; Greenberg
1986
; Rhode 1995
). However, the monotonically rising input-output functions of auditory-nerve fibers represent a
further problem. An increase in discharge rate must inevitably lead to
shorter ISIs, thus interfering with an analysis of these intervals in
terms of quasi-periodicity associated with the pitch sensation. One
possible way to overcome this problem is the processing of higher-order
ISIs, an operation equivalent to an autocorrelation of the spike train
(Cariani and Delgutte 1996a
,b
; Shofner
1991
, 1999
). A stimulus periodicity encoded in
first-order ISIs at low stimulus levels may be preserved in
higher-order ISIs for higher sound levels. This was confirmed
experimentally by Cariani and Delgutte (1996a
,b
), who
found that a neural correlate of pitch in the cat auditory nerve is
well preserved in an all-order ISI analysis, whereas a first-order
analysis was susceptible to changes in sound level. Whereas this is
certainly true for the temporal encoding of pitch in the auditory
nerve, this conclusion cannot readily be extended to the cochlear
nucleus. In the cochlear nucleus, the neural information provided by
the auditory nerve is subjected to different types of temporal and,
through the interaction of units with different "best frequencies"
(BFs), spectral processing. Concerning temporal processing, onset units
were found to accentuate the degree of AM of an acoustic stimulus
(Frisina et al. 1990a
,b
; Kim and Leonard
1988
; Kim et al. 1986
; Rhode and
Greenberg 1994
; Wang and Sachs 1994
). Chopper
neurons also accentuate AM, but they do so only for a limited range of
modulation frequencies that are in the vicinity of their chopping
frequency. Thus chopper neurons have temporal modulation transfer
functions (a measure of temporal discharge synchrony as a function of
modulation frequency) with a band-pass characteristic (Kim et
al. 1990
). This distinguishes chopper neurons from primary-like
(PL) units or auditory-nerve fibers that typically show modulation
transfer functions with a low-pass characteristic and a relatively
small degree of temporal synchronization to a sinusoidal modulator.
In most previous studies on temporal neural correlates of pitch,
research focused on the correlation between periodic envelope modulation, as seen in the stimulus waveform, and the periodicity of
neural discharge. However, some stimuli do not reveal a pronounced envelope modulation in the stimulus waveform, but they nonetheless elicit a clear pitch perception. Examples for these stimuli are random-phase or Schroeder-phase harmonic complexes (the latter are
designed to minimize waveform modulation) or iterated rippled noise
(IRN) (Wiegrebe and Patterson 1999; Yost
1996a
,b
).
Rippled noise is generated by adding a delayed copy of a sample of white noise back to itself. Iterating this delay-and-add process generates IRN. In spectral terms, IRN is characterized by a ripple spectrum that, with increasing number of iterations, approximates the spectrum of a harmonic complex. In temporal terms, the iterated delay-and-add process introduces some degree of temporal periodicity that is reflected by a peak in the normalized autocorrelation function of the stimulus at a correlation lag equal to the IRN delay. This peak grows with increasing number of iterations. The height of the peak is a measure of the degree of periodicity of the stimulus. As this peak never reaches a value of one (perfect periodicity) with increasing number of iterations, we use the term quasi-periodicity when describing IRN stimuli. Perceptually, IRN produces a pitch sensation with a pitch corresponding to the reciprocal of the delay and a pitch strength that grows with increasing number of iterations. An example of an IRN waveform with a delay of 8 ms and 16 iterations is shown in the top panel of Fig. 1 and is compared with an example of a white-noise waveform (bottom).
|
In a physiological study using rippled noise, Shofner
(1991) showed in the chinchilla cochlear nucleus that the
rippled-noise delay is well represented in the all-order ISI statistics
of PL units. Shofner (1999)
confirmed these findings
using infinitely iterated rippled noise, i.e., a stimulus type with a
much stronger pitch than rippled noise. However, Rhode
(1995)
using a variety of pitch-producing stimuli, showed that
cat cochlear nucleus units represent pitch in their first-order ISI characteristics.
To investigate the representation of the pitch of IRN, we have examined
the responses of single units in the guinea pig cochlear nucleus in
terms of all-order and first-order ISIs (Winter et al.
1999). Sustained-chopper neurons as well as OC neurons revealed a band-pass temporal tuning based on first-order ISI statistics. This
was not only the case for pitch stimuli with highly modulated envelopes
such as cosine-phase harmonic complexes but also for pitch stimuli with
relatively flat envelopes like random-phase complexes and IRN. While
the strong temporal representation of the stimulus envelope by OC units
has been well documented by several groups (Kim and Leonard
1988
; Kim et al. 1986
; Palmer and Winter
1992
, 1993
; Rhode 1994
,
1995
; Rhode and Greenberg 1994
), to our
knowledge the band-pass periodicity selectivity of the OC units had not
been previously reported. In addition the OC units showed a range of
periodicity preferences to fundamental periods as long as 16 ms
(Winter et al. 1999
).
In this study, the influence of presentation level is investigated systematically. As argued above, a temporal pitch code based on first-order ISIs could be highly sensitive to changes in presentation level because an increasing level may lead to an increase in discharge rate and thus to decreasing ISIs. This may be more the case for a stimulus with a relatively flat envelope like IRN (cf. Fig. 1, top) than for a stimulus with pronounced modulation like modulated pure tones or a cosine-phase harmonic complex because, due to the flat envelope, firing times can be more evenly distributed.
Given these stimulus features, the encoding of the quasi-periodicity of
IRN stimuli over an appreciable range of delays and presentation levels
represents a significant challenge for a mechanism of temporal-pitch
extraction based on first-order ISIs in the cochlear nucleus
(Rhode 1995). Nevertheless, the current results suggest
that first-order ISIs of some units in the cochlear nucleus can provide
a reliable representation of the pitch of IRN over a wide range of
levels. Specifically, sigmoidally saturating, sustained-chopper units
show a band-pass temporal tuning that is relatively stable over a 30- to 50-dB level range. The results suggest that a complex, higher-order
temporal code of stimulus quasi-periodicity, as it is preserved in the
auditory nerve and in PL cochlear nucleus neurons (Cariani and
Delgutte 1996a
,b
; Shofner 1999
), may be
converted into a first-order temporal code by chopper neurons of the
cochlear nucleus. This could be achieved through a facilitatory
interaction of the complex periodicity at the neuron's input with the
neuron's intrinsic oscillation. In the range of stimulus levels where
the rate response saturates, the first-order temporal code provides a
reliable, level-independent estimate of the stimulus pitch.
![]() |
METHODS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Preparation
The data reported in this paper were recorded from 14 pigmented guinea pigs weighing between 319 and 471 g. Animals were anesthetized with urethan (1.5 g/kg ip) and supplementary analgesia provided by either operidine (1 mg/kg im) or fentanyl (1 mg/kg im). All animals were given Atropine sulfate (0.06 mg/kg sc) as a premedication. Additional doses of urethan and the analgesic were given when required.
The surgical preparation and stimulus presentation took place in a
sound-attenuating chamber (IAC). All animals were tracheotomized and
core temperature maintained at 38°C with a heating blanket. Following
placement in the stereotaxic apparatus, a midline incision of the scalp
was made and the skin retracted laterally. The temporalis muscle on the
left-hand side of the skull was removed and the bulla exposed. The
method of stereotaxic positioning follows that previously reported
(Winter and Palmer 1990). The stereotaxic coordinates
were identical to those used in previous studies in the ventral and
anteroventral cochlear nucleus (Winter and Palmer 1990
,
1995
), and electrode tracks sometimes coursed their way through the dorsal cochlear nucleus before entering the ventral division. This transition was marked by a change in best frequency of
single units. Although data were recorded from units in the dorsal
cochlear nucleus, as judged by their stereotaxic position and
physiological response type, we have excluded them from the present
data set. No histological verification of recording position was
undertaken, but the above observations indicate that all the units
reported in this paper were recorded from the ventral division of the
cochlear nucleus.
The compound action potential (CAP) was monitored with the use of a
silver-coated wire placed on the round window of the cochlea. The
signal was filtered and amplified (×10,000). The CAP threshold was
determined visually (10-ms tone-pip, 1-ms rise/fall time, 10 s1) at selected
frequencies at intervals during the experiment. If thresholds had
deteriorated by more than 10 dB and were not recoverable (for example,
by removal of fluid from the bulla), the animal was killed by an
anesthetic overdose of pentobarbitol sodium (ip).
Recording technique
Recordings were made using tungsten-in-glass microelectrodes
(Merrill and Ainsworth 1972). Electrodes were
advanced by an electronic microdrive (Kopf 650W) through the intact
cerebellum in the sagittal plane at an angle of 45°. A wideband noise
stimulus was used to locate the surface of the cochlear nucleus and to search for single units.
Stimuli
For the isolation of single units, we used 50-ms bursts of
broadband noise at a repetition rate of 4 Hz. For the classification of
single units, stimuli were 50-ms pure tones at the unit BF. They were
generated at a sampling rate of 200 kHz and low-pass filtered at 40 kHz. The complex stimuli consisted of white noise or IRN with 11 different delays ranging from 1 to 32 ms in half-octave steps, a gain
of 1 and 16 iterations. IRN was generated by delaying and adding white
noise. When this process was iterated, the output waveform of one
delay-and-add stage served as the input to the next stage ["add
same" configuration in Yost (1996a,b
)]. The IRN stimuli were generated digitally at a sampling rate of 20 kHz using the
Tucker Davis System II DSP Board and software delay lines. The stimulus
duration was 409.6 ms including 10-ms cosine-squared ramps. Due to the
nondeterministic nature of white noise and IRN, 25 samples were
generated off-line for each delay and stored to hard disk. The digital
stimulus energy was kept constant. To assess the effect of stimulus
level, white noise and IRNs with 1 of 11 delays were presented at 7 different levels in 5-dB steps that amounted to a total of 84 different
stimulus conditions. In a few examples, the effect of level was
assessed over a 50-dB range corresponding to a total of 132 conditions.
Each of these conditions was presented 25 times using the 25 different
waveforms. For the data collection, the 84 or 132 conditions were
presented in a randomized order that, however, was the same for each of
the 25 repetitions. The stimuli were presented at a repetition rate of 1 per s, which resulted in an overall measurement duration of 35 or 55 min for the 30-dB range or the 50-dB range, respectively. After D/A
conversion, the complex stimuli were low-pass filtered at the Nyquist
frequency (10 kHz, TDT FT6) and attenuated (TDT PA4). The stimuli were
equalized (Phonic PEQ 3600 Equalizer) to compensate for the speaker-
and coupler frequency response before being fed into a Rotel RB971
power amplifier and a custom-made programmable end attenuator (0-75 dB
in 5-dB steps). The different stimulus attenuations were set on the
PA4; the minimal attenuation required for a specific measurement was
set on the end attenuator to optimize the signal-to-noise ratio. The
signal was presented over a Radio Shack speaker mounted in a coupler
designed for the guinea pig ear (Mike Ravicz, MIT). The stimuli were
acoustically monitored with a B&K 4134 microphone attached to a
calibrated, 1-mm, probe tube. For an attenuation of zero, a maximum
sound-pressure level of 122 dB SPL for the broadband stimuli was
possible; however, it should be noted that this level was never used,
and the vast majority of recordings were limited to maximum levels of
102 dB SPL. The frequency response of the set-up was flat between 100 Hz and 10 kHz within ±3dB. The 30- or 50-dB range of attenuations presented to a specific unit was usually set to be just above threshold
for the highest stimulus attenuation.
Analyses
UNIT ISOLATION AND CLASSIFICATION.
On isolation of a unit, the BF was determined audio-visually using
pure-tone stimulation of variable frequency and level. Units were
classified based on their BF peri-stimulus time histograms (PSTH), the
first-order ISI histogram (ISIH) and a regularity analysis
(Young et al. 1988). PSTHs were obtained for 250 presentations of the 50-ms BF tones (including 1-ms raised-cosine
ramps) at a repetition rate of 4 Hz presented at levels of 20 and 50 dB above unit threshold. Spontaneous rate was determined in a 10-s time
window before the PSTHs were measured. Chopper units were subdivided
into sustained choppers [CS, coefficient of variance (CV)
0.35] or transient choppers (CT, CV > 0.35) (see Young et
al. 1988
).
RESPONSE TO IRN AND WHITE NOISE. To quantify the extent to which the quasi-periodicity of the IRN stimuli is reflected in the temporal response characteristics of a unit, first-order and all-order ISIHs were calculated. To determine the influence of stimulus quasi-periodicity on the neural response, ISIHs in response to IRN stimuli were compared with the ISIHs in response to white-noise, and a so-called "interval enhancement" value was calculated: the proportion of intervals with a duration equal to the IRN delay was reduced by the proportion of the same-duration intervals in response to white noise. For example, when a unit was stimulated with IRN with a delay of 4 ms, and 25% of all ISIs were 4 ms, and, with white-noise stimulation, only 20% of the intervals were 4 ms, the interval enhancement was 5%.
To compare temporal response properties between white-noise and IRN stimulation, the white-noise, first-order ISIHs, and the IRN interval-enhancement as a function of IRN delay were fitted with a gamma function of the form
![]() |
(1) |
![]() |
RESULTS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
This study is based on recordings from a total of 145 neurons; from this group at least 1 complete delay-level map was obtained for 54 units.
CS units
First-order ISIHs of the responses of a CS unit with a BF of 1.25 kHz to IRN stimuli with various delays and attenuations are shown in Fig. 2. The figure only contains the data for half of the IRN delays actually presented; responses for the intermediate delays of 1.4, 2.8, 5.6, 11.2, and 22.4 ms were omitted for reasons of clarity. In response to white noise (WN), the first-order ISIH shows the typical chopper characteristic with a peak at an interval corresponding to the chopping period. In response to IRN with a 4-ms delay, this peak is strongly enhanced compared with the white-noise response. Figure 3, A and B, shows the interval enhancement for this unit as a function of IRN delay and level. Figure 3A shows the interval enhancement based on the first-order ISIHs; Fig. 3B shows the interval enhancement based on the all-order ISIHs. Positive values indicate that the ISIH for IRN contains more intervals corresponding to the IRN delay than the white-noise ISIH. Based on first-order statistics only (Fig. 3A), the unit reveals interval enhancement with a band-pass characteristic. Moreover, this band-pass interval enhancement is relatively level-independent with a peak at 4 ms over a range of 40 dB. Based on an all-order ISIH analysis (Fig. 3B), the band-pass characteristic is lost, and the neuron shows interval enhancement for all delays higher than 2.8 ms (long-pass). Figure 3, C and D, shows interval enhancement for another CS unit with a BF of 0.85 kHz. As in the previous example, the unit shows band-pass interval enhancement. The best delay is at 2.8 ms, and it shifts very little over the 30-dB range tested.
|
|
CT units
CT units are distinguished from CS units based on their higher CV,
which indicates that the regularity of the temporal discharge is not as
high as that of CS units and the mean ISI in response to pure tones
typically rises over the stimulus duration. CT units also differ from
CS units in terms of the nonmonotonicity of their input-output
functions (Blackburn and Sachs 1990; Winter and
Palmer 1990
). Whereas CS units show a steep rise over an about
20- to 25-dB range and then frequently saturate, the input-output
function of CT units often show nonmonotonic shapes. Interval
enhancement as a function of IRN delay and level for a CT unit with a
BF of 1.1 kHz (Fig. 4, A and
B) and a unit with a BF of 1.35 kHz (Fig. 4, C
and D) are plotted in Fig. 4. The left and
right panels present interval enhancement based on
first-order and all-order statistics, respectively. Compared to the CS
units shown in the previous figures, a band-pass characteristic of the
interval enhancement is less pronounced in these CT units despite their
similar BF. Moreover, due to the typically wider dynamic range,
interval enhancement based on first-order ISIs is more susceptible to
changes in presentation level.
|
PL units
PL units reveal frequency tuning and temporal discharge
patterns similar to those of auditory-nerve fibers (e.g., Bourk
1976; Rhode and Smith 1986
; Winter and
Palmer 1990
). ISIHs for IRN responses of a PL unit with a BF of
0.75 kHz are shown in Fig. 5 in the same
format as in Fig. 2. In response to white noise of different levels
(1st column), the ISIH for this PL unit shows the typical monotonic decay. In response to IRN stimuli, peaks in the ISIHs corresponding to the IRN delay are found for a broad range of delays
between 2 and 8 ms.
|
Interval enhancement for first-order ISIs and mean-rate responses as a function of IRN delay and level for two PL units are plotted in Fig. 6. Figure 6, A and B, shows responses of the unit from Fig. 5 with a BF of 0.75 kHz; Fig. 6, C and D, shows responses of a unit with a BF of 2.47 kHz. For both PL units, interval enhancement based on first-order interval statistics (left) does not show such a pronounced band-pass characteristic as seen for the CS units (Fig. 3). Moreover, strong level effects are observed. For the low-BF unit, the interval enhancement decreases with increasing stimulus level for the whole range of delays where the unit reveals some degree of interval enhancement at the lowest stimulus level (attenuation = 60 dB). The loss of interval enhancement with increasing level could be explained by inhibitory input to this PL unit. The presence of inhibition can be visualized by plotting the mean-rate response as a function of IRN delay and sound level in the right panels of Fig. 6. While the 2.47-kHz unit (Fig. 6D) shows a monotonically increasing rate response with increasing level, the 0.75-kHz unit (Fig. 6B) shows a decrease in the mean-rate response with increasing level. Thus it is likely that this unit receives inhibitory as well as excitatory input. The high-BF unit does not show these inhibitory effects.
|
Onset units
The PSTH of onset units is characterized by a high probability of
spike discharge at stimulus onset. The tonic response can differ in
strength and temporal characteristics. For the current analyses of the
temporal discharge characteristics in response to IRN stimuli, an
additional inspection of the first-order ISIH of the BF-tone response
provides a more reliable means to separate different types of onset
units than the PSTH alone. In this analysis, we distinguish between
onset-chopper units (OC) and onset units with low-level sustained
response (OL) (Godfrey et al. 1975). OC and OL units are
typically distinguished based on the BF-tone PSTH. The OC PSTH shows a
second (and rarely a 3rd) sharp peak following the onset peak, whereas
additional peaks are absent in OL PSTHs (Rhode and Smith
1986
; Winter and Palmer 1995
). Examples of onset
unit responses to pure-tone and IRN stimulation are given in Fig.
7. An OC unit is shown in the left
column, and an OL unit is in the right column. The
top row shows PSTHs, the middle row shows the
corresponding first-order ISIHs, and the bottom row shows
interval enhancement as a function of IRN delay and sound attenuation.
The PSTH of the OC unit (Fig. 7A) shows a pronounced second
peak about 2 ms after the first peak. The PSTH of the OL unit (Fig.
7B) does not show a second peak but a gap following the
onset peak in its PSTH. The ISIHs of the units' pure tone responses
(Fig. 7, C and D) reveal a chopping of both these
neurons in the sustained response and not only the chopping of the OC unit at stimulus onset. The interval enhancement plots (Fig. 7, E and F) reveal a band-pass characteristic not
unlike that described for CS units (Fig. 3). As in the CS units, the
band-pass characteristic appears to be related to the chopping in the
pure-tone, sustained response.
|
A unit that would be classified as Onset according to the
classification scheme used by Winter and Palmer (1995)
is shown in Fig. 8. They used the ratio
between the magnitude of the onset response and the sustained response,
which had to be more than 10 to 1 for a unit to be classified as onset.
The PSTH of the unit in Fig. 8A meets this criterion, but
the ISIH (Fig. 8B) shows that the unit reveals no chopping
in the sustained response but an ISIH similar to a PL unit. Thus this
unit could also be classified as a primary-like unit with a notch (PN).
A clear distinction between OL and PN units can be difficult based on
the PSTH (e.g., Rhode 1994
). In our sample of units, OL
units all show a chopping in the pure-tone sustained response. Thus it
is suggested that the ISIH may provide a means to distinguish between
OL and PN units.
|
The amount of interval enhancement in response to IRN appears to be related to the shape of the first-order ISIH in response to equal-level white noise. This is illustrated in Fig. 9, where we plotted the position of the interval-enhancement peak as a function of the position of the peak in the white-noise, first-order ISIH. To obtain a reliable estimate of relation between white-noise and IRN stimulation, a gamma function was fitted to both the white-noise ISIH and the interval-enhancement functions as described in METHODS. Each data point in Fig. 9 is derived from an iso-level recording of the white-noise response and the responses to IRN with 11 different delays and the same level as the white noise. Thus up to seven data points (for 7 different levels) were derived from a single unit. Figure 9 shows that there is a reasonable correlation (R = 0.79) between the peak in the noise ISIH and the position of the interval enhancement peak. This is true despite the fact that, to obtain the interval enhancement, the noise ISIH was subtracted from the IRN ISIHs.
|
The use of a gamma function to fit the noise ISIH enables us to correlate changes in the fit parameters, n and b, with the unit type from which noise responses were obtained. For this analysis, units with band-pass characteristics as opposed to units with PL, low-pass, characteristics are pooled, i.e., CS, CT, and OC as opposed to PL and PN units. The average values and standard errors for the order parameter, n (Fig. 10A) and the bandwidth parameter, b (Fig. 10B) are plotted for these two coarse classifications in Fig. 10. Again, the number of analyses, N, represents the number of different noise ISIHs obtained (which can be up to 7 for the 7 different levels in 1 unit). Chopping units are characterized by a higher value for n and b. This reflects the occurrence of a mode in the noise ISIH, i.e., a delayed and steep rise followed by an similarly steep drop in the noise ISIH. For units with a PL sustained response, the low values for n and b characterize a noise ISIH with an immediate, steep rise, mostly due to neural refractoriness and a decay that is well fitted as an exponential with a short time constant, b.
|
To summarize the effect of presentation level, neural responses from OC units are compared with responses of CS units. These two unit types are interesting to compare because both show interval enhancement with a band-pass characteristic suitable to code the IRN pitch in first-order ISIs, but they differ in their dynamic ranges. As the interval-enhancement peak position is correlated with the noise ISIH peak position (cf. Fig. 9), the noise ISIH peak position is plotted as a function of the stimulus attenuation in Fig. 11 for nine CS units (Fig. 11A) and six OC units (Fig. 11B). In general, both unit types show a trend toward an earlier peak position with increasing stimulus level. However, this trend appears to be confined to low presentation levels (attenuations higher than 60 dB corresponding to sound levels lower than 62 dB SPL) for the CS units (Fig. 11A) above which the peak appears level independent. OC units show significant IRN responses only at attenuations lower than 45-40 dB (sound levels higher than about 80 dB SPL). The shift in the peak position with increasing level is more pronounced for OC units (Fig. 11B).
|
![]() |
DISCUSSION |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
This study has investigated the temporal representation of IRN in the guinea pig cochlear nucleus as a function of delay and presentation level. Temporal response characteristics of cochlear-nucleus units were analyzed in terms of both first-order and all-order ISIs. Whereas interval enhancement showed a low-pass characteristic for all neuron types in all-order ISIHs, a pronounced band-pass characteristic of interval enhancement was found for both CS and OC units in first-order ISIHs. Moreover, especially for CS units, this interval enhancement appeared to be rather insensitive to changes of presentation level. The peak of the band-pass interval enhancement of the CS units shown in Figs. 2 and 3 changes over the range of levels where the rate response increases. This is the case because an increasing spike rate with increasing level does necessarily lead to shorter ISIs. When driven into saturation, however, the units' temporal firing characteristics still reveal the temporal quasi-periodicity of the IRN stimulus. This results in a stable, first-order temporal pitch code that is level independent in the saturated range. This finding supports the idea that CS units may play an important role in the pitch perception associated with complex harmonic sounds that is not susceptible to level changes.
Responses of CT units to IRN with varying delays and presentation
levels differ from those of CS units in that the responses are
typically more susceptible to changes in presentation level and that
the interval enhancement appears not to have such a pronounced band-pass characteristic. These two differences can be related to
response features with pure-tone stimulation; for example the decrease
in discharge regularity as a function of poststimulus time and
nonmonotonicity in BF rate-level functions (Blackburn and Sachs
1989; Winter and Palmer 1990
).
OC and OL units have a wide dynamic range. As long as a unit's rate
response is not yet saturated, the peak of the interval-enhancement functions shifts toward shorter delays with increasing sound level (cf.
Fig. 7E). This is the case because first-order ISIs
inevitably become shorter when the rate response increases. Assuming a
first-order ISI code for pitch in the cochlear nucleus, OC and OL units
like CS units may provide a conversion of higher-order to first-order intervals, but the wide dynamic range of OC and OL units makes them
susceptible to changes in presentation level. Whereas it is believed
that CS units project to the inferior colliculus (IC) (Adams
1979; Smith et al. 1993
), projection sites of OC
units are still unclear, and it may be possible that they act as
interneurons in the CN (Joris and Smith 1998
).
Comparison with previous studies
Shofner (1991) investigated the temporal
representation of rippled noise (RN) in the antero-ventral
cochlear nucleus of the chinchilla. He concluded that, while PL units
seemed to preserve the RN fine structure, chopper units only code the
quasi-periodicity in the stimulus envelope. This was based on the
finding that rippled noise that was delayed and added (the gain,
g, in the delay-and-add loop equals 1) was coded in the same
way as rippled noise that was delayed and subtracted (g =
1). Shofner (1999)
investigated the temporal
response characteristics of chinchilla cochlear nucleus units in
response to infinitely iterated rippled noise (IIRN) with a
g of 0.89 or
0.89 in the delay-and-add loop. As for the rippled-noise results, Shofner (1999)
concluded that
while PL units do preserve the difference between IIRN with positive
and negative g, this was not the case for chopper units.
IIRN with positive and negative g share the same envelope
features but differ in their temporal fine structure (Yost et
al. 1998
). This argument would be consistent with the idea that
chopper units are envelope responders and PL units are driven by
fine-structure information. Shofner (1999)
showed
responses of two PL units with BFs of 0.85 and 4.63 kHz. Whereas the
autocorrelation of the low-BF response reflected the stimulus
autocorrelation, the high-BF units' autocorrelation was the same
irrespective of the sign of the IIRN gain as it was the case for the
2.43-kHz chopper unit shown in Shofner (1999)
. Thus it
is possible that the encoding of the IIRN temporal properties is more
dominated by an effect of BF than by an effect of unit type. This
interpretation would also be more in line with the perception of the
stimuli. For a fixed delay of 4 ms, the pitch difference between
positive and negative g is an octave only when the low
harmonics are presented. When the stimuli are high-pass filtered, the
pitch difference is much smaller and more on the order of 10%
(Wiegrebe, unpublished data) as it is observed for rippled noise
(Bilsen 1966
; Bilsen and Ritsma
1969/1970
; Yost et al. 1978
). It is therefore
possible that chopper units with low BFs may well be capable of
preserving differences related to the sign of g in their
temporal response properties as far as these are established
perceptually. This issue clearly needs to be explored in more detail.
Comparison with studies on amplitude-modulated (AM) tones
Frisina et al. (1990a,b
) investigated the encoding
of AM pure tones in the gerbil cochlear nucleus. They calculated the
modulation gain based on the PSTH of the unit response to AM tones of
various modulation frequencies and sound levels. The modulation gain, a
measure to quantify the synchrony of the firing to the modulator, revealed a low-pass characteristic at low stimulus levels. At higher
stimulus levels, chopper units revealed a band-pass characteristic of
the modulation gain not unlike the band-pass characteristic of interval
enhancement found in the present study with IRN stimulation. In this
section, we compare the two measures of synchrony [the PSTH-based
measure of Frisina et al. (1990a
,b
) and the ISIH-based measure used here] and show data from a single CS unit where we obtained responses to both AM tones and IRN. Frisina et al.
(1990a
,b
) determined the modulation gain in the PSTH by
dividing the percent modulation of the stimulus with the percent
modulation of the PSTH. The latter was calculated by dividing the
Fourier component of the PSTH at the modulation frequency by the
average of all Fourier components. As the degree of modulation in an
IRN stimulus is not known, one cannot calculate the modulation gain for
an IRN stimulus. Instead, we directly compare the Fourier component at
the modulation frequency of the PSTHs in response to AM tones and IRN.
Figure 12A shows the height
of the Fourier component of the PSTHs for a CS unit with a BF of 0.99 kHz for AM-tone (- - -) and IRN stimulation (
). The function is
low-pass for AM-tone stimulation with a peak at 350 Hz, which indicates
a possible transition to a band-pass characteristic when the sound
level would be increased, as it was described in Frisina et al.
(1990a
,b
). With IRN stimulation, the height of the Fourier
component is very low and shows no systematic dependence on modulation
frequency (equals 1/delay). The same data analyzed in terms of interval enhancement are plotted in Fig. 12B: with this type of
analysis, both data sets reveal a clear band-pass characteristic of the temporal tuning properties of the unit. The PSTH-based measure does not
work for IRN stimulation because IRN is a nondeterministic stimulus
that, in our experiments, was refreshed for each presentation. Thus to
the extent that IRN will cause modulation in a frequency channel, the
modulation will probably have a different phase with every
presentation. Thus the PSTH, which is summed up over all presentations,
does not reveal a possible modulation.
|
First-order ISI characteristics are obviously not confounded with these
effects. The comparison of the two analysis types shows that it is not
straightforward to compare the data by Frisina et al.
(1990a,b
) to our data, but the general finding of a band-pass characteristic of the temporal tuning can be found with both AM-tone and IRN stimulation. Frisina et al. (1990a
,b
) described
that the modulation-gain functions for CS units change from low-pass to band-pass with increasing stimulus level. In principal, this can be
confirmed with IRN stimulation. First-order interval enhancement for
the CS unit in Fig. 3A reveals a broader tuning with a peak shifted toward longer delays at low presentation levels. With increasing level, the band-pass characteristic becomes more pronounced, and the peak becomes fixed at 4 ms.
Possible implications for the modeling of pitch processing in the auditory brain stem
Like PL units, chopper units in the cochlear nucleus preserve the
quasi-periodicity information of IRN stimuli with positive gain in the
all-order ISIHs. Chopper neurons, however, can be interpreted as not
only preserving the quasi-periodicity information but as the first
stage of a temporal processing that renders the need for an
autocorrelation, i.e., an all-order ISI analysis unwarranted. Chopping
produces a decrease in the number of intervals that do not correspond
to the chopping period and an increase in the number of intervals equal
to the chopping period. If this chopping period equals the
stimulusquasi-periodicity, the chopper-unit output is strongly
locked to the stimulus period even when the stimulus period is not
reflected in pronounced periodic envelope oscillations of the stimulus.
In an array of chopper units with the same BF but with a range of
chopping periods as hypothesized by Frisina et al.
(1990a,b
) and Kim et al. (1990)
, the stimulus
quasi-periodicity would be represented in an interval place code.
Hewitt and Meddis (1994)
demonstrated in a computer
model of AM sensitivity of single units in the inferior colliculus how
such a first-order interval-place code can be converted into a
rate-place code in coincidence detector units presumably located in the
central part of the inferior colliculus. The current experiments
indicate that the suggestions of Hewitt and Meddis
(1994)
for the coding of AM pure tones could be extended to the
case of more complex periodic stimuli where the periodicity is not
apparent in the stimulus envelope. In this suggested circuitry, the VCN
CS units could not only provide the all-order to first-order conversion
but also a level insensitivity of temporal periodicity coding. Due to
the relatively high sensitivity and steeply rising input-output
functions of CS units with a range of only 20-25 dB (Blackburn
and Sachs 1989
; Rhode and Smith 1986
;
Winter and Palmer 1990
), their rate response saturates
at relatively low levels. Above this saturation level, the first-order
interval code is level independent.
It should be recognized that this hypothesis requires a two-dimensional
representation of BF and periodicity tuning (Frisina et al.
1990a,b
; Kim et al. 1990
). However, to date only
one study has shown a distribution of best periodicity as a function of BF (see Kim et al. 1990
, Fig. 12). They found best
periodicities ranging from 100 to 500 Hz in a population of units in
the posteroventral and dorsal cochlear nucleus. No single unit type was
found to encompass the whole range of best periodicities corresponding to the range of pitches perceived. Interestingly, the range of best
periodicities in chopper units varied between 90 and 400 Hz.
While the data reported here do not encompass the range of BFs and
temporal periodicities necessary to verify a two-dimensional representation of the two, the current results nevertheless provide physiological evidence in support of Rhode's (1995)
suggestion that periodicity pitch in the cochlear nucleus is encoded in
first-order ISIs. In contrast to the auditory nerve, this code can be
insensitive to changes in presentation level.
![]() |
ACKNOWLEDGMENTS |
---|
We thank K. Krumbholz and G. Neuweiler for critical comments on earlier versions of this paper. R. Patterson has continually provided fruitful discussions on the perception and encoding of iterated rippled noise.
This study was supported by a research grant from the Deutsche Forschungsgemeinschaft to L. Wiegrebe, the Medical Research Council, and the Wellcome Trust, UK.
![]() |
FOOTNOTES |
---|
Present address and address for reprint requests: L. Wiegrebe, Zoologisches Institut der Universität München, Luisenstr. 14, 80333 Munich, Germany (E-mail: wiegrebe{at}zi.biologie.uni-muenchen.de).
Received 7 July 2000; accepted in final form 21 November 2000.
![]() |
REFERENCES |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|