Signal Processor for and an Auditory Prosthesis Having Spectral to Temporal Transformation

The present invention relates generally to auditory prostheses and signal generators for auditory prostheses. More particularly, the present invention relates to such signal processors employing a plurality of band pass filters.

Various types of multichannel auditory prosthetic devices exist in the art.

One example of a multichannel auditory prosthesis is described in U.S. Pat. No. 4,400,590, to Michelson. Michelson discloses a system that uses the theory that differing places along the cochlea respond as differing frequencies to the brain. Thus, band pass filters break up the incoming auditory signal into a plurality of frequency bands. These signals are then applied directly to electrode locations along the cochlea. In theory, those locations which correspond to its associated frequency band.

Another example of a multichannel auditory prosthetic device, is described in U.S. Pat. No. 4,289,935, to Zollner et al and U.S. Pat. No. 4,403,118, to Zollner et al. The system described in the Zollner et al patents uses a set of band pass filters to generate frequency bands to turn on and off, or to modulate, oscillators (tone generators) whose outputs are then summed and transmitted to a hearing aid.

Both of the Michealson and Zollner et al multichannel auditory prosthetic devices, however, have been less than completely successful in obtaining open set speech comprehension without the use of visual aids.

Young and Sachs, in a 1979 article, "Representation of Steady-State Vowels in the Temporal Aspects of the Discharge Patterns of Auditory Nerve Fibers", 66 Journal of the Acoustic Society of America, pp 1381-1403, noted that spectral information is represented in the timing patterns of auditory nerve discharges. They determined the energy at a given frequency by examining temporal responses only among neurons whose center frequencies were close to that frequency.

The present invention operates on the principle that individual nerve cells within the auditory central nervous system responds only to appropriate periodicities, i.e., only those periodicities close to the inverse of the center frequency for each invididual auditory nerve fiber.

The present invention provides a signal processor for an auditory prosthesis which has a plurality of highly tuned band pass filters coupled to an electrical input signal which is designed to represent sound to a person. Each of the plurality of filters is less than critically damped, thus, each of the plurality of filters has an impulse response which is oscillatory. In a preferred embodiment, the less than critically damped feature is achieved by a band pass filter which is a Q (3 dB) which is more than 0.5. In a still preferred embodiment, each individual band pass filter has a Q which corresponds to the average Q of auditory nerves which respond to the same frequency as the center frequency of each of the plurality of filters. The output of each individual band pass filter may optionally be passed through a nonlinearity device, preferably compressing the signal, and then may either be individually supplied to a multielectrode stimulating system or partially or completely summed with each other and supplied to a single electrode pair stimulating system or a multielectrode pair stimulating system in which there are fewer electrode pairs than there are filters.

The highly tuned band pass filters of the present invention will create output signals from each individual band pass filter which contain the periodicities corresponding to the frequency component in the input signal corresponding to the center frequency of the band pass filter. Thus, a stimulation signal with the appropriate periodicity of a particular center frequency will cause the auditory central nervous system to respond with a signal which represents a signal to the person of that center frequency even though such signal is delivered to many places along the cochlea provided the signal is delivered to the place along cochlea whose neurons are normally tuned to that center frequency.

Thus, the auditory prosthesis, and signal processor, of the present invention obtain several properties of a normal cochlear function. First, the temporal behavior of each band pass filter mimics that of an individual neuron tuned to the same frequency of that band pass filter. Second, the higher the center frequency of the band pass filter, the faster is its transient response. Thus, a transient stimulus will elicit a response first from the highest frequency band pass filter, then from the second highest, and so on until a response occurs from the lowest frequency band pass filter. This aspect mimics the propagation of the response up the basilar membrane from the base to the apex, duplicating the normal propagation delay of the cochlea. Third, the phase response of each band pass filter can be adjusted to match the phase response of a neuron at the corresponding place along the cochlea. This produces a phase vs. place profile which mimics that of a normal cochlea. Thus, normal velocity vs. place stimulation will be obtained with slowing in the vicinity of the best frequency. Fourth, the Bekesy traveling wave envelope along the basilar membrane is approximated by the band pass filter outputs, each representing the basilar membrane motion at a place corresponding to its center frequency.

The foregoing advantages operation and construction of the present invention will become more readily apparent from the following figures in which:

FIG. 1 is a simplified block diagram of the signal processor of the present invention;

FIG. 2 is a representation of the frequency spectrum of the vowel "eh";

FIG. 3 illustrates a transfer function of the band pass filters;

FIG. 4 illustrates the time domain band pass filter outputs with a stimulus as described by FIG. 2;

FIG. 5 illustrates the amplitude characteristics of a preferred band pass filter;

FIG. 6 illustrates the phase characteristic of a preferred band pass filter;

FIG. 7 illustrates the actual Q's of neurons for various frequencies plotted as a function of center frequency;

FIG. 8 is block diagram of a preferred embodiment of the auditory prosthesis of the present invention;

FIG. 9 is an exemplary input/output function of a nonlinearity;

FIG. 10 is a flow chart illustrating how the nonlinearity of FIG. 9 is defined; and

FIG. 11 is a block diagram of an alternative embodiment of the present invention.

Work by Young and Sachs has demonstrated that spectral information is well represented in the timing patterns of auditory nerve discharges. In their analysis, they determined the energy at a given frequency by examining temporal responses only among neurons whose center frequencies were close to that frequency.

The present invention recognizes that the auditory central nervous system may perform a similar analysis, ignoring inappropriate periodicities, in responding only to appropriate ones, i.e., those close to the inverse of the center frequency for each individual nerve fiber. This ability to ignore an inappropriate periodicity on a nerve fiber can effectively explain the very good speech recognition achievable with single electrode cochlear implants. Periodic events corresponding to both fundamental and formant frequencies may exist in a single electrode stimulus waveform. With appropriate equalization, each of these periodicities may appear in the temporal discharge patterns of neurons across the entire cochlea. However, they will be ignored by the central nervous system except when they occur at the corresponding appropriate sites along the cochlea. For example, a complex stimulus signal containing both 500 Hertz and 200 Hertz components includes some events in the time domain which occur at 2 millisecond intervals and some which occur at 5 millisecond intervals. Such a stimulus signal delivered electrically to the cochlea would elicit neuronal impulses at both 2 millisecond and 5 millisecond intervals. These impulses would be elicted everywhere along the cochlea but the two millisecond intervals would be ignored except from the 500 Hertz neurons. Similarly, the 5 millisecond intervals would be ignored except from the 200 Hertz neurons. This illustrates that a single stimulating electrode, stimulating many neurons, can nevertheless deliver information about more than one frequency.

However, it is recognized that the number of different periodicities which may be contained on a single electrode may be limited. For example, if a single electrode attempts to convey a large number of differing periodicities, the combination may create additional periodicities not present in the original signal which could confuse the auditory central nervous system. For example, if 50 different periodicites were occuring on a particular neuron, the appropriate ones, e.g., 0.5 millisecond for a two kiloHertz fiber, might occur too rarely to be detected by the nerve fiber. This limitation can be overcome by the use of multiple electrodes each of which stimulates a spatially distinct group of neurons. It would then be unnecessary for one electrode, or electrode pair, to deliver all of the periodicities contained in the incoming signal. Rather, each electrode, or electrode pair, would deliver only intervals corresponding to the center frequencies of the fibers within its locus or spatial range of excitation. This system has the potential to increase dramatically the frequency resolution afforded by multiple electrode auditory prosthetic devices which depend on spacial selectivity for achieving frequency differentiation.

A simplified block diagram of a signal processor to achieve the appropriate periodicities discussed above is illustrated in FIG. 1. The signal processor 10 receives an electrical input signal 12. The electrical input signal 12 is supplied to a plurality of sharply tuned band pass filters 14, 16 and 18. These band pass filters 14, 16 and 18 are less than critically damped and provide an electrical output signal 20, 22 and 24. These band pass filters 14, 16 and 18 have an impulse response which is oscillatory. In a preferred implementation, each of the band pass filters 14, 16, and 18 has a Q (3 dB) of more than 0.5. The sharply tuned band pass filters 14, 16 and 18 operate as resonators so that the output signals 20, 22 and 24, respectively, contain periodicities corresponding to the energy in the electrical input signal 12 corresponding to the center frequency to which band pass filters 14, 16 and 18 are tuned. Output signals 20, 22 and 24 may, optionally, be routed through nonlinearities 26, 28 and 30, respectively. Nonlinearities 26, 28 and 30 may operate to compress output signals 20, 22 and 24 so that the dynamic range of the signal present may be fitted into the dynamic range remaining in the person to be stimulated. In one embodiment, the outputs of nonlinearities 26, 28 and 30 are then summed in summer 32 and a single output signal 34 is provided which contains the periodicites present in electrical input signal 12. Output signal 34 may then be provided to the remainder of an auditory prosthesis to be delivered to an electrode to electrically stimulate the person or may be provided to an electrical-to-auditory transducer in order to acoustically stimulate a person. While the block diagram of the signal processor in FIG. 1 illustrates three separate band pass filters, resonators, 14, 16 and 18 it is to be recognized and understood that more resonators or fewer resonators may be desirable in a particular signal processor and that three are illustrated for illustrative purposes only.

Since too many periodicites present in output signal 34 may confuse the individual neurons of the auditory central nervous system, it may be preferred that only those output signals 20, 22 and 24 which are largest in amplitude be applied to the auditory central nervous system and, hence, be contained in output signal 34. In the signal processor of FIG. 1, each band pass filter, resonator, 14, 16 and 18 is a very narrow filter whose purpose is to generate an output signal 20, 22 and 24 which is periodic if the electrical input signal 12 has significant energy near its center frequency. The period of that signal is the inverse of the frequency being passed through the resonator 14, 16 and 18. The nonlinearity 26, 28 and 30 may have a "deadband" which serves to mask out low level signals. Signals appear at the outputs of only those nonlinearities 26, 28 and 30 whose input signals 20, 22 and 24 exceed the deadband threshold. The nonlinearities 26, 28 and 30 do not alter the periodicities contained in those signals. The nonlinearities 26, 28 and 30 merely determine whether they are large enough to be passed on and to limit them from becoming too large, i.e., compression. As the signals from nonlinearities 26, 28 and 30 are summed in summer 32, a composite output signal 34 is provided. Note that only a few of signals 20, 22 and 24 may be actually present in output signal 34 because only those signals 20, 22 and 24 which have sufficient amplitude to exceed the deadband of nonlinearities 26, 28 and 30 actually appear at summer 32. Thus, output signal 34 may have more than one periodicity but not so many periodicities that the periodicities are lost.

FIG. 2 shows the frequency spectrum of a vowel "eh" 36, before appropriate high frequency emphasis 38 at 30 dB per octave. The spectrum of the vowel "eh" 36 shows formant peaks near 0.5 kiloHertz and 2.0 kiloHertz. The frequency spectrum of the vowel illustrated in FIG. 2 is then applied to each of ten resonators 40 illustrated in FIG. 3 tuned at one-half octave intervals between 0.177 kiloHertz and 4.0 kiloHertz. The transfer function of resonators 40 correspond to resonators 14, 16 and 18 of a signal processor as illustrated in FIG. 1. The time domain output signals of these resonators 40 are illustrated in FIG. 4 by reference numerals 42-60. Notice that the time domain output signals 54 and 46 are the greatest in amplitude which corresponds to the 0.5 kiloHertz and 2.0 kiloHertz formant peaks of the frequency spectrum of FIG. 2. Thus, resonators 40 have created periodicites, mainly signals 54 and 46, which corresponds to the frequency of energy contained in the input signal of the vowel 36.

The amplitude characteristic 62 of a preferred resonator is illustrated in FIG. 5. Correspondingly, the phase characteristic 64 of a preferred resonator is illustrated in FIG. 6. The sharpness of each reasonator, or band pass filter, is measured by the Q of the reasonator. The Q is determined by dividing the center frequency of the resonator by the bandwidth of the resonator as measured by the frequencies which are given amount dB down from the peak of the resonator response. Thus, the value of Q must be expressed in terms of Q for a particular dB. If the bandwidth of the resonator response as measured 3 dB down from the peak is a given frequency width and that frequency width is divided into the center frequency, the Q(3 dB) is obtained. Similarly, a Q(10 dB) can be obtained by taking the bandwidth at the 10 dB down points on the amplitude curve 62 of FIG. 5. Thus, it can be seen that the higher the value of Q the more sharply tuned the individual resonator. Also, it can be seen that a given resonator will have a Q(3 dB) which is of a greater numerical value than its corresponding Q(10 dB). For purposes of the present invention, the resonator response must be underdamped, that is, the Q(3 dB) must be more than 0.5. The resonator must have an oscillatory response to a step or impulse input. This is referred to its impulse function being oscillatory. As illustrated in FIG. 6, the phase characteristic of the reasonator is positive for frequencies below the center frequency of the resonator, has a sharp transition through zero at the center frequency and is negative for frequencies greater than the center frequency of the resonator. This mimics the phase characteristic of the individual neurons of the auditory central nervous system.

Ideally, in a preferred embodiment, the Q of each individual reasonator should match the Q actually measured for nerve fibers with similar center frequencies. Such Q's have been measured by Kiang et al "Discharge Patterns of Single Fibers in the Cat's Auditory Nerve", MIT Research Monograph 35, Library of Congress 6614345 (1965). The Q's 66, as determined by Kiang et al, are depicted graphically in FIG. 7. Here the Q's 66 illustrated are Q(10 dB) values.

FIG. 8 represents a block diagram of a preferred embodiment of a auditory prosthesis 68 of the present invention. A microphone 70 transforms a given auditory signal into an electrical input signal 12. The electrical input signal 12 is fed to a switchable lowcut filter 72 to optionally provide noise supression for low frequency noise components. An automatic gain control 74, which may have an external sensitivity control, limits the dynamic range of the electrical input signal 12. A pre-emphasis filter 76 boosts the high frequency components common in speech signals. A gain element, amplifier, 78 compensates for internal signal losses. An anti-aliasing filter 80 prevents corruption of the signal by frequencies above the Nyquist frequency of 5 kiloHertz. Another gain element, amplifier, 82 again compensates for internal losses. A sample and hold circuit 84 and analog to digital converter element 86 convert the signal to a digital representation. Elements 70, 72, 74, 76, 78, 80, 82, 84 and 86 are conventional and well known in the art. Elements 72, 74, 76, 78, 80, 82 and 84 are optional and are illustrated here as being part of the preferred embodiment of the auditory prosthesis 68.

A digital signal processor 10 corresponds to the signal processor 10 illustrated in FIG. 1. Signal processor 10 contains a plurality, namely nine, band pass filters 88A-88I, or resonators, each passing a different center of frequency. In a preferred embodiment, resonator 88A has a center frequency of 0.35 kiloHertz; resonator 88B, 0.5 kiloHertz; resonator 88C, 0.71 kiloHertz; resonator 88D, 0.91 kiloHertz; resonator 88E, 1.17 kiloHertz; resonator 88F, 1.5 kiloHertz; resonator 88G, 1.94 kiloHertz; resonator 88H, 2.5 kiloHertz; and resonator 88I, 3.2 kiloHertz. The individual outputs of resonators 88A-88I are passed to nonlinearities 90A-90I. The outputs of these nine nonlinearities 90A-90I are then provided to summer 32 where the signals are digitally summed together and converted back into an analog signal by digital to analog converter 92. An inverse filter 94 is provided to compensate for any subsequent known output transmission characteristics of the auditory prosthesis. An internal volume control 96 is supplied to allow for appropriate amplitude adjustment by the person and the signal is then provided to an electrode, or electrode pair, 98. Elements 92, 94, 96 and 98 are conventional and well known. Elements 94 and 96 are optional. In a preferred embodiment, microphone 70 is a Knowles EA1934 with 3 dB downpoints at 250 hertz and 8 kilohertz. In a preferred embodiment, lowcut filter 72 is a 6 dB per octave lowcut filter with corner frequencies switch selectable by the patient at either 250 hertz or 500 hertz. In a preferred embodiment, automatic gain control circuit 74, has an attack time of approximately 1 millisecond and a release time of approximately 2 seconds. The threshold is determined by a sensitivity control. In a preferred embodiment, pre-emphasis filter 76 is a 6 dB per octave high pass filter with a corner frequency at 4 kilohertz. Pre-emphasis filter 76 is intended to compensate partially for the 10-12 dB per octave high frequency roll off in the long term spectrum of speech, thus, decreasing the loss of amplitude resolution in high frequency components of speech. Gain element 78, anti-aliasing filter 80, gain element 82, sample and hold element 84, analog to digital converter element 86, digital to analog converter element 92, inverse filter 94, volume control 96 and electrode 98 are conventional elements well known in the art. In the preferred embodiment of the auditory prosthesis 68, resonators 88A-88I are digitally implemented at their indicated center frequencies. In a preferred embodiment, resonator 88A and resonator 88B have a Q(10 dB) equal to 3, resonators 88C, 88D and 88E have a Q(10 dB) equal to 4, resonator 88F has a Q(10 dB) equal to 5, resonator 88G has a Q(10 dB) equal to 6, resonator 88H has a Q(10 dB) equal to 7 and resonator 88I has a Q(10 dB) equal to 8. The digital implementation of all filters are preferred to be fourth order. The digital implementation of such filters are conventional in design and are well known although not used for this purpose or function.

The implementation of nonlinearities 90A-90I is a function of input fitting and output fitting. Each nonlinearity maps a range of instantaneous input signal levels as produced by resonators 88A-88I, respectively into a range of instantaneous output signal levels. The mapping function has a linear region 100, a cascade of two power functions 102 and 104 and a saturation function 106 as shown in FIG. 9. By mapping the input levels into the saturation region 106 or into either of the power function regions 102 and 104, any range of input levels can be compressed into the desired output range. The output range is tailored to the subject's electrical dynamic range. The input range is also set as part of the input fitting to map a desired range of resonator output levels into the compressive region of the output range and thus into the subject's electrical dynamic range.

The output range is fitted to the subject with three nonlinearity parameters Y(min) 108, Y(mid) 110 and Y(max) 112. Y(min) 108 defines the boundary between the linear function 100 and the first power function 102. Y(min) 108 is set at the subject's perceptual threshold, i.e., the level below which the subject has no auditory perception. Y(max) 112 defines the boundary between the second power function 104 and the saturation function 106. Thus, output levels greater than Y(max) 112 are not produced. Y(max) 112 for an individual subject is determined in conjunction with the uncomfortable loudness level of the individual subject. Y(mid) 110 which corresponds to the boundary between the first power function 102 and the second power functions 104 is defined as the value of Y(min) 108 plus 0.66 times the quantity Y(max) 112 minus Y(min) 108.

The input dynamic range of each nonlinearity 90A-90I is fitted to the distribution of the instantaneous signal levels in speech measured at the output of the individual resonators 88A-88I. Three parameters X(min) 114, X(mid) 116 and X(max) 118 which define corresponding coordinate pairs with the Y values 108, 110 and 112 are used to fit the nonlinearity. Two optional approaches in fitting the X values 114, 116 and 118 can be utilized.

In a first approach, the so called standard approach, the 95th percentile of the distribution instantaneous output levels from each resonator 88 was computed for a large sample of processed speech. X(max) 118 is set to this value. Using this value, 5% of the resonator output for an individual channel is mapped into the saturation function 106. X(min) 114 is then set to a value 20 dB below X(max) 118 and X(mid) 116 is set halfway between X(max) 118 and X(min) 114. These parameters map approximately 40%-50% of the input speech levels into the subject dynamic range for each channel. With the remaining 50%-60% falling at or below threshold in the linear function 100, this approach mixes a large number of channels which contain suprathreshold components. Thus, the composite signal from summer 32 may exhibit a large number of periodicities.

In a preferred embodiment, a "channel dominance" approach is utilized in which only the outputs from resonators 88A-88I whose signals are likely to be the largest pass or are gated through nonlinearities 90A-90I and, thus, are gated onto summer 32 and, hence, whose periodicities are present in the output signal which are above the patient's perceived threshold level. In this way, the number of periodicities present in the summed signal are reduced and appropriate possible periodicity interactions are also reduced. This is accomplished by increasing the values of the X parameters 114, 116 and 118 so that a greater percentage of the speech input signals would be mapped into the linear function 100 and, thus, be below the perceptual level of the patient. Utilized in this manner, only the highest amplitude or dominant channels are at suprathreshold levels. The 75th percentile was identified from the instantaneous level distribution for an individual channel and was utilized as the value for X(min) 114. X(max) 118 was set 20 dB above X(min) 114 and X(mid) 116 was set 10 dB above X(min) 114.

With the X values 114, 116 and 118 and the Y values 108, 110 and 112 determined as specified above, a flow chart for a program to digitally implement the function of the nonlinearities 90 is illustrated in FIG. 10. Digitial program operates by starting at block 200 and individually implementing sequentially each individual channel A-I. Since each individual channel is identical to the other, only one channel implementation, namely channel A will be described. A digital filter or resonator is implemented by block 202 which corresponds to resonator 88A. This digital implementation is conventional and well known in the art. For the first channel, the program in block 204 takes the ouuput from block 202 and implements the linear function 100 of FIG. 9 by determining if the X value is within the range of the linear function 100. If yes, the program determines the output function Y(t) by obtaining constant values A1 and B1 and applying them to the formula A1.times.X(t)+B1. If the value of X(t) is not within the range of the linear function 100, then block 206 determines if the X values are within the range of the first power function 102. If yes, then the program determines the Y output value by looking up values A2 and B2 in a table and applying it to the formula Y(t)=A2.times.X(t)+B2. Preferably this can be accomplished by a plurality of piecewise linear segments to approximate alogrithmic curve. Similarly, the second power function 104 is implemented in block 208, if appropriate by using the table look up for the values A3 and B3. Also, preferably alogrithmic curve may be approximated with linear segments. Finally, the program implements the saturation function 106 by block 212 if the X value X(t) is within the value of the saturation portion by simply outputting a known constant since the curve is saturated. The program then repeats each of these individual blocks for each individual channel contained within the digital signal processor 10.

In the preferred embodiment, the program illustrated in FIG. 10 is implemented on an integrated circuit which forms and operates as signal processor 10. This integrated circuit is a model 320C10 integrated circuit manufactured by Texas Instrument Corporation, Dallas, Tex. In the program of FIG. 10, the value obtained from each individual channel is then digitally summed at block 214 to obtain the final Y(t) and the program is ended at block 216 and, hence, repeats.

An alternative embodiment of the auditory prosthesis 120 is illustrated in FIG. 11. The auditory prosthesis 120 is similar to the auditory prosthesis 68 of FIG. 8 in that it has a microphone 70 and automatic gain control circuit 74, a pre-emphasis of filter 76 and a signal processor 10. Signal processor 10 contains a single A to D converter 86 which supplies the electrical input signal to 10 resonators 88A-88J. The center frequencies at 178 Hertz, 250 Hertz, 353 Hertz, 500 Hertz, 707 Hertz, 1 kiloHertz, 1.4 kiloHertz, 2 kiloHertz, 2.8 kiloHertz and 4 kiloHertz, respectively. Each resonator 88 is supplied to a nonlinearity 90A-90J individually. Resonators 88 and nonlinearities 90 operate identical to the resonators 88 and nonlinearities 90 illustrated in FIG. 8. However, in the embodiment illustrated in FIG. 11 the output of the individual nonlinearities instead of being summed together are individually passed to digital to analog converters 92A-92J. Alternatively, one digital to analog converter 92 could be coupled to all nonlinearities 90A-90J which could handle all ten nonlinearities by being time division multiplexed. The output of each individual digital to analog converter 92A-92J is supplied to an individual current source 122A-122J which individually supply an electrode 124A-124J. Thus, each individual resonator 88A-88J ultimately supply an individual electrode 124A-124J. Thus, an individual electrode, e.g., 124A, would contain only the periodicities of one resonator, namely, resonator 88A. The wires supplying electrodes 124A-124J are illustrated passing through a percutaneous plug 126. Electrodes 124A-124J while illustrated in FIG. 11 as a single wire will exist in the patient as an electrode pair since they must pass current from one individual electrode element to a second individual electrode element. Thus, electrode 124A may represent a wire element pair designed to be positioned within the cochlea. Alternatively, electrodes 124A-124J may represent single electrode elements which pass current from their individual elements to a single common return electrode (not shown).

While the description of the preferred embodiment is described as electrically stimulating a person as in a cochlear implant, it is to be recognized and understood that the present invention is also applicable to acoustic stimulation as in a hearing aid.

While the description of the preferred embodiment is described as having the predetermined characteristics of the nonlinearities and/or filters as being set for all time, it is to be recognized and understood that these predetermined values could advantageously be recomputed occasionally, periodically or upon the occurrence of a certain event and still be predetermined in the intervals in between.

Thus, there has been shown and described a novel signal processor for an auditory prosthesis having spectral to temporal transformation. It is to be recognized and understood, however, that various changes, modifications and substitutions in the form and of the details of the present invention may be made by those skilled in the art without departing from the scope of the following claims: