[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: mechanical cochlear model

Just for the record, I did not side with the travelling wave hypothesis.

I did want to say that there appears to be a divergence on what 'travelling wave' means. It would be good if someone knowledgeable gave an authoritative definition. If possible make it operational by saying how we can recognize that there is one, and how we would rule out the opposite hypothesis.

In the physics courses I took way back when, travelling waves were introduced in the context of propagation on a string or in a waveguide. Since the cochlea differs considerably from a uniform string or waveguide it's not immediately obvious what a travelling wave is in this context. The physics are quite different. In particular the fact that propagation is quasi-instantaneous in the liquid surrounding the membrane makes the system very different from a string (or the surface of the sea).

The 'simple-account-for-students-in-cognitive-sciences' didn't mention a travelling wave.


I agree with Reinhart Frosch and Alain de Cheveigne and Dave Mountain on this. The cochlear traveling wave may have some open issues, but to say that "its function in hearing is not yet universally appreciated" is misleading, and I think a bit disingenuous.

It's OK to investigate the open issues, and to use them to motivate new ideas that depart from the generally accepted view. But I think it's not a good idea to misrepresent the extent to which the ideas are accepted by the mainstream hearing community. In the case of the traveling wave, the acceptance is pretty much universal, and has been for quite a long time; the agreement of theory and experiment gets better over time, as experimental data get better and as analysis techniques get more mature and models converge on physical measurements.

With respect to Jont Allen's 2001 remark that "the discrepancy in frequency selectivity between basilar membrane and neural responses has always been, and still is, the most serious problem for the cochlear modelling community" (in his chapter "Nonlinear Cochlear Siganl Processing" in Jahn and Santos-Sacchi "Physiology of the Ear"), I had encountered that myself recently, and was wondering what's behind it. Jont goes on to say, in italics even, "In my view, this discrepancy is one of the most basic unsolved problems of cochlear modeling." and then "Progress on this front has been seriously confounded by the uncertainty in, and the interpretation of, the experimental data."

It's a great chapter, and I agree with Jont on many of his points and attitudes, but I still wonder what he was poking at with that paragraph. My view is a little different: active hydromechanical models are able to yield transfer functions and tuning curves and nonlinearities that are pretty much like what we see in both mechanical and neural data (FTCs, Weiner kernels, 2TS curves, etc.). The problem may be in his intepretation of "frequency selectivity", as he sometimes turns FTCs upside-down and compares them with transfer functions, as in his Figure 19-12 that he refers back to later as an example of a neural/mechanical mismatch. That seems like an odd mistake to make in a chapter on nonlinearity, but it is one that is widespread in the hearing literature, not unique to Jont, and one that has a history of introducing confusion about sharpness in different ways of looking at a system.

Jont also says, a few pages earlier, "In fact, according to measurements made over the last 20 years, the response of the basilar membrane to a pure tone can change in amplitude by more than 5 orders of magnitude per millimeter of distance along the basilar membrane (ie, 300 dB/octave is equivalent to 100 dB/mm in the cat cochlea)." This strikes me as another misapplication of linear thinking to a nonlinear system. If the FTC requires an increase of 5 orders of magnitude to get a given response when the frequency is increased by 1/3 octave, which I can accept as roughly credible, then that has to be understood as a combination of a fairly sharp high-frequency rolloff and a fairly strong nonlinear level dependence. If the response compression slope is about 1/3 above CF, then the compression explains about 2/3 of the steepness of the 300 dB/octave slope, so the estimated change in response with place for a given pure tone needs to be cut by a factor of three, to about 1.7 orders of magnitude, not 5. This is the kind of discrepancy that I'd think would be included when saying "Progress on this front has been seriously confounded by the uncertainty in, and the interpretation of, the experimental data," but perhaps Jont is thinking of some other problems.

The open problems that Jont points out, of getting good models of two-tone suppression and upward spread of masking, that unify mechanical and neural data, are indeed important points where modelers have more work to do. I think the situation is actually in not such bad shape, though. I don't see any real basis for thinking that a new paradigm is what's needed at this time to make progress, and I don't think that's what Jont is saying, either, unless he's telling us to stop assuming that the system has active amplification.

Jont, can you fill us in? Has the problem been updated in your opinion since 2001? Are you referring to interpretations of active amplification as part of the problem?