RECORDING CLICK-EVOKED OTOACOUSTIC EMISSIONS

         USING MAXIMUM LENGTH SEQUENCES 

(Summary of Thornton et al., 1994) 

ARD Thornton

MRC Institute of Hearing Research, Royal South Hants Hospital

Southampton, Hants, SO14 0YG

Telephone: (023) 8063 7946, Facsimile: (023) 8082 5611

Email: ardt@soton.ac.uk

1.      INTRODUCTION

Neonatal hearing screening, using evoked otoacoustic emissions (EOAEs) would be improved if the testing could be speeded up and made more sensitive to detect the small responses that occur shortly after birth (Kennedy et al. 1998, Thornton 1999). It was to address these two problems that the feasibility of applying maximum length sequence (MLS) techniques to evoked emissions was investigated.  The duration of the evoked emission is  of  the order of 20 ms so if the speed of the technique is increased by simply increasing the click presentation rate, the responses to successive clicks will start to overlap each other at rates greater than about 50 clicks/s, the rate recommended by Kemp et al. (1990).  It would be impossible to  recover the normal, evoked emission from these overlapped recordings.  However, if a particular sequence of clicks and silences, known as a maximum length sequence (MLS), is presented then  the overlapped responses can be deconvolved to give the original response that would have been obtained from conventional, slow, averaging.  Currently, such a technique has been applied to otoacoustic emissions with click rates of up to 5000 clicks/s.

2.      MAXIMUM LENGTH SEQUENCES

FIG.1

For each stimulation sequence of 1s and 0s a recovery sequence can be obtained by replacing each 0 in the stimulus sequence with -1.  An MLS and its recovery sequence are shown in Figure 1. Details of MLS generation and deconvolution have been published (Davies, 1966;  Burkhard  et al.,  1990) and  the first audiological application of  MLS  was given by Eysholdt and Schreiner (1982).

One way of visualising the recovery process is shown in Figure 2 using an MLS of length 3  comprising the stimulus sequence 1, 1, 0; the corresponding recovery sequence being  1, 1, -1.  To perform the recovery, the stimulus sequence is  rotated left by the minimum inter-stimulus interval two times to complete the matrix shown on the left hand side and is then multiplied by the  recovery sequence. 

FIG. 2

When the right hand matrix, containing the multiplied values, is summed the  ‘recovered' stimulus is obtained at twice its original amplitude (because the MLS stimulus  sequence had two clicks in it) with all other, later elements being cancelled to zero.  In order to improve the signal-to-noise ratio (SNR), time domain averaging is carried out as normal, with  analogue-to-digital converter (ADC) samples for corresponding points in consecutive  presentations of the MLS being summed.  For the MLS recovery process to work, there must be no gap between the deconvolved MLSs that will be added to the average.

This means that there must be no additional delay between ADC samples when the presentation of one  MLS finishes and the next begins; the time between samples being typically of the order of 30 µs.

In conventional averaging, the normal way of rejecting records that are too noisy is to have a reject level criterion that comes into effect when the click stimulus has passed.  The problem with MLS recordings, particularly at the higher rates, is that the click stimuli and emission responses  overlap, with the responses 'riding on top' of the stimuli and so it is simply not possible to properly reject noisy response epochs with such MLS stimuli. 

3.      "ON-THE-FLY RECOVERY"

To overcome this problem a procedure has been developed and named "on-the-fly recovery." Instead of adding each incoming ADC sample to a summation buffer the new method multiplies each sample as it arrives by the values in the recovery sequence and adds the results directly into the appropriate positions in a recovery buffer.  As soon as the last sample for one MLS has been dealt with, the recovery buffer contains the recovered response from all the stimuli in that MLS. By using double-buffering, each recovered response can be checked against the rejection criteria and accepted or rejected while the response to the next presentation of the MLS builds up in a second buffer; accepted responses are added to a final summation buffer. 

FIG.3

Consider an MLS of length 3 with stimulus sequence 1, 1, 0.  The 3 slices of the MLS are represented by M1, M2 and M3 (Figure 3[A]).  The recovery process, seen before, is illustrated again in Figure 3[B] and, in terms of the slices, in Figure 3[C].

An additional benefit of the "on-the-fly recovery" method is that it can recover only those parts of the MLS that are of interest.  The set of recovery sequence points to be used will change as one works through the MLS, but in a systematic way, so that finding the correct set is computationally simple.  This method allows the recovery window to be positioned anywhere within the duration of the MLS provided that the edges of the window correspond to points in the MLS; that is when a stimulus opportunity of either a click or a silence is occurring.  This is illustrated in Figure 3[D].

Recovery windows starting and ending at intervening points can obviously be achieved by rounding up the actual recovery window used to a whole number of MLS points and then subsequently discarding the unwanted portions. Figure 3[E] illustrates the recovery procedure using the ADC samples which are labelled a to i.  The position of each sample in the final buffer is shown in Figure 3[F].

4.      DYNAMIC RANGE

Since the small response overlaps and rides on top of the large click stimuli, a large dynamic range and good linearity are needed for MLS systems (Thornton, 1993b).  The dynamic range needed has been estimated as 94dB (Thornton et al., 1994).  Noise in the system is the limitation of the SNR and there are three noise sources identified so far :-

1.      Random noise from the microphone, amplifiers and ADC quantisation.  The largest contribution to the noise component is from the microphone.  Our system has at best about a 70 dB signal to noise ratio so, in the raw input signal, the EOAE is only approximately 20 dB above the noise floor.  However because these sources of noise are random they are reduced by averaging, the standard formula of 10.log (n) dB improvement in SNR occurs for n  averages.  Furthermore, if this noise is less than the noise from the test room and from subject or patient movement, then it will have little or no effect on the averaged waveform.

2.      Noise due to non-linearities in the system.  If the system is driven into a non-linear region of operation, the reconstructed waveform will have noise that is produced as a result of that non-linearity and since it is synchronised to the clicks, it will not be reduced by averaging.  Thus all distortion products should be less than the required 94 dB.

3.      Noise due to incomplete cancellation.  As explained above, the MLS technique works on the basis of cancellation, i.e. when one click waveform is subtracted from another, the result must be exactly zero.  If the clicks differ by only 1%, a 0.5% residual (the difference between each click and the mean click) would be left, giving a dynamic range of only 46 dB.  It is therefore very important that the clicks are matched precisely in both time and amplitude.  Non-random variations of click amplitude will not be reduced by averaging.

5.   SYSTEM DYNAMIC RANGE AND LINEARITY

FIG. 4

At high stimulus rates the stimuli become very close to each other. At 3000 clicks/s they are beginning to overlap each other.  At 5000 clicks/s the overlap is considerable.  This can be seen in Figure 4 which shows stimuli generated in an IEC 2 cc cavity in which a B&K microphone was mounted to record the signal.  The 5000 clicks/s stimulation rate does not look anything like a train of clicks as the stimuli have very nearly merged together.  However, there are two points of note. 

Firstly the ear, as will be shown later, appears to respond to this as a set of clicks presented at a rate of 5000/s.  Secondly the deconvolution procedure works for both the stimulus as well as the response and this fact enables us to check our Institute's in-house MLS system to see if any non-linearities have altered the clicks and caused interaction between the stimuli.

FIG. 5

Figure  5 shows the deconvolved click stimuli, albeit recorded through the system's 500 to 5000 Hz bandpass filter and therefore broadened somewhat, obtained at rates from 40 to 5000 clicks/s.  There appears to be very little difference between the conventional click at 40/s and the MLS clicks that follow. 

FIG. 6

Figure 6 shows the same data with the separation between click waveforms reduced to zero.  This enables the fine detail of the structural changes to be seen and it is clear that such differences as there are can be seen only as a slight increase in the line width that occurs from about 1.5 ms onwards.  The initial part of the click waveform, the one probably responsible for generating the EOAE, varies virtually not at all over stimulus rate. 

This indicates that both the dynamic range of the recording system and its linearity are good enough if the stimuli can be deconvolved to this degree of accuracy. 

5.      RESULTS

  Figure 7 shows the waveforms for evoked emissions recorded conventionally and with the MLS technique at stimulus rates up to 5000 clicks/s. It can be seen that there is a decrease in the long latency, low frequency portion of the OAE waveform as stimulus rate increases.  This rate effect and other normative properties have been detailed elsewhere (Hine and Thornton, 1997).  Happily, for applications in neonatal screening, the short latency, high frequency part of the emission is much less altered by this rate effect.

The responses recorded at stimulus rates for which the clicks are clear distinct events, do not show any major changes from the responses recorded at the highest rates for which the clicks merged together.  Thus, as mentioned earlier, the auditory system appears to be responding to the ‘merged’ clicks in the same way as it does to the ‘distinct’ ones.  The technique has also been used to investigate pathological conditions (Hine et al, 1997; Norman et al, 1996) and applied to neonates (Slaven and Thornton, 1998).

There are applications to neonatal screening because, for neonates with good OAEs, the MLS technique can pass the baby some 13 times faster than the conventional technique.  However, the most important aspect of using MLSs is that, if averaging is done for the same time as the conventional technique, then responses can be detected that are only 20% of the amplitude of those that would be detected by the conventional response (Hine et al., 2001).  Given the small responses obtained on Day 1 of a neonate’s life this could be important in the future.

There are other areas of investigation which the MLS technique makes possible particularly those involving recording the non-linear temporal interaction responses generated by the cochlea (Thornton, 1997; Thornton et al, 2001).  Our current work indicates that these non-linear components may be more sensitive to pathology than the conventional ones.

 REFERENCES

Burkard R, Shi Y, Hecox KE.  Brainstem auditory evoked responses elicited  by maximum length sequences; Effects  of  simultaneous masking noise.  J Acoust Soc Am  1990; 87: 1656-64.

Davies  WDT.   Generation  and  properties  of  maximum   length sequences.  Control 1966; 10: 364-5.

Eyesholdt U, Schreiner C .  Maximum length sequences – a fast method for measuring brainstem evoked responses.  Audiology 1982; 21: 242-50.

Hine , J.E. and Thornton, A.R.D. Transient evoked otoacoustic emissions recorded using maximum length sequences as a function of stimulus rate and level.  Ear and Hearing, 1997, 18, 121-128.

Hine JE, Thornton ARD, Brookes GB.  Effect of olivocochlear bundle section on evoked otoacoustic emissions recorded using maximum length sequences.  Hearing Research 1997; 108:28-36.

Hine JE, Ho CT, Slaven A, Thornton ARD.  Comparison of transient evoked otoacoustic emission thresholds recorded conventionally and using maximum length sequences.  Hearing Research 2001; 156:104-14.

Kemp  DT,  Ryan  S, Bray P.  A guide to  the  effective  use  of otoacoustic emissions.  Ear Hear 1990; 11: 93-105.

Kennedy CR, Kimm L, Cafarelli Dees D, Campbell MJ, Thornton ARD.  A controlled trial of universal neonatal screening for early identification of permanent childhood hearing impairment.  The Lancet 1998; 352: 1957-64.

Norman M, Thornton ARD, Slaven A, Phillips AJ.  Otoacoustic emissions recorded at high rates in patients with confirmed acoustic neuromas.  American Journal of Otology 1996; 17:763-72.

Slaven, A and Thornton, ARD.  Neonatal otoacoustic emissions recorded using maximum length seguence stimuli.  Ear & Hearing 1998; 19(2): 103-110.

Thornton ARD.  High rate otoacoustic emissions.  J Acoust Soc Am 1993; 94: 132-6.

Thornton ARD, Folkard TJ, Chambers JD.  Technical aspects of recording evoked otoacoustic emissions using maximum length sequences. Scandinavian Audiology 1994; 23: 225-31.

Thornton ARD.  Maximum length sequences and Volterra series in the analysis of transient evoked otoacoustic emissions.  British Journal of Audiology 1997; 31:493-8.

Thornton ARD.  Maturation of click evoked otoacoustic emissions in the first few days of life. In: Grandori F, Collet L, Ravazzani P, editors. Otoacoustic Emissions from Maturation to Ageing. Series in Audiology Number 1. London: Decker Europe ; 1999. p. 21-32.

Thornton ARD.  Temporal non-linearities of the cochlear amplifier revealed by maximum length sequence stimulation.  Clinical Neurophysiology 2001; 112: 768-777.