betaFeedback

WorksPhysics and Astronomy

Multiple spontaneous otoacoustic emissions as a single standing wave unstable to modulation

Abstract

Spontaneous otoacoustic emissions (SOAEs) have been interpreted either as cochlear standing waves or as local oscillations of hair cells. Multiple SOAE lines test both views, because each view must explain how the lines coexist. In the standing-wave view, each line is a separate resonance. In the intermodulation view, some lines are distortion products of others. Both explanations assume several independent oscillations. Independent oscillations, however, compete through the saturating cochlear amplifier, and the dominant oscillation suppresses the others. The saturation has another role besides suppression. The saturation also induces an oscillation of the gain at twice the frequency of the oscillation. In lasers, this oscillation of the gain, called the pump, creates sidebands. This paper examines whether a cochlear standing wave can pump the gain by itself. A two-dimensional cochlear model with a step in the gain produced a single standing wave. The pump destabilized a pair of lines near the neighboring resonances. The other lines were phase-locked intermodulation products of the main line and the pair. A group of equally spaced SOAEs can thus arise from one self-sustained oscillation. The spacing follows the resonances of the standing wave, and the two explanations can hold together.

Introduction

Spontaneous otoacoustic emissions (SOAEs) are narrow-band sounds that the ear emits without stimulation[1], and the origin of SOAEs is not settled[2]. In one view, mammalian SOAEs are amplitude-stabilized cochlear standing waves[3]. Waves reflected within the cochlea and at the stapes build up at the frequencies where the round-trip phase is a multiple of 2π2\pi and the round-trip gain reaches unity. The saturation of the cochlear amplifier fixes their amplitude[4, 5]. In the other view, SOAEs come from critical (Hopf) oscillators of the hair cells[6, 7]. Coupled arrays of such oscillators reproduce the statistics of human SOAEs[8]. Human SOAEs have also been modeled as the synchronized clusters of a chain of coupled van der Pol oscillators[9]. Spectra with several lines can help to decide, because each view must explain how the lines coexist.

Multiple SOAEs in one ear have been explained in two ways, as separate resonances or as intermodulation products. In the standing-wave view, each SOAE is a separate resonance. Neighboring SOAEs are then spaced by about the inverse of the round-trip delay[3, 10]. The same delay sets the period of the fine structure of stimulus-frequency OAEs (SFOAEs) and of the microstructure of hearing thresholds[11, 12]. In the other view, some SOAEs are intermodulation products of others. In triplets of equally spaced SOAEs, the phases of the three peaks lock as those of a cubic distortion product[13]. Such triplets are the exception, since the ratios of the frequency differences of consecutive SOAEs generally do not cluster at unity, the value expected for intermodulation[14]. The two explanations thus differ in whether every SOAE is an independent oscillation.

Both explanations assume several independent oscillations, and independent oscillations tend not to coexist. Independent oscillations share the saturating cochlear amplifier, and a strong component suppresses the others, as in two-tone suppression[15]. In cochlear models, the dominant oscillation suppresses the other modes[16]. The standing-wave view must therefore explain why several resonances oscillate at once. The intermodulation view must likewise explain why two or more primary oscillations survive. Locked phases alone do not reveal the origin of a line, because coupled independent oscillators also lock[9, 17]. In cochlear models, the lines other than the dominant one have been attributed to independent unstable modes or to their distortion products[16, 18]. If a single standing wave generated the other lines, the lines would coexist without competing. The spacing would then follow the resonances, while the other lines would be intermodulation products. The two views would hold together.

Analyses of cochlear models have treated the saturation only as an average drop of the gain. The eigenvalues of a model at rest show which modes start to grow[19, 20]. Once an oscillation saturates the amplifier, the saturation has been included as a gain averaged over a cycle[16]. The averaged gain is lower than the gain at rest. A lower gain can only suppress the other lines, not create new lines.

The saturation, however, also induces an oscillation of the gain, and the oscillating gain can create sidebands. An oscillation saturates the gain at each peak, twice in each cycle. The gain therefore oscillates at twice the frequency of the oscillation. This oscillation of the gain is called the pump here. In lasers, an oscillating gain creates a pair of lines on both sides of the main line[21, 22]. There, the gain responds to the intensity averaged over the optical cycle, and it oscillates at the beat frequency of the main line and a sideband. In the cochlear amplifier, the gain follows the displacement itself, and it oscillates at twice the frequency even for a single line. Both oscillating gains couple the two sidebands by four-wave mixing. In ears, sidebands of an SOAE have been produced by an external low-frequency tone[23]. A cochlear standing wave may thus pump the gain by itself and create lines near the neighboring resonances.

This paper shows that a group of equally spaced SOAE lines can arise from a single cochlear standing wave that the pump renders unstable to modulation. The mechanism concerns such a group from one standing wave, not SOAE spectra in general (§ 7 B).

The model is a two-dimensional (2D) cochlea with a lumped middle ear[24]. Unlike distributed roughness[4, 25], a step in the gain places the reflection at a single known position[26], so the resonances are known. The eigenvalues of the model about the standing wave, computed with the pump, show which lines the pump destabilizes. The time-domain solution confirms the eigenvalues.

Model

Cochlear fluid

Figure 1 shows the model, which consists of a probe in the ear canal, the middle ear, the cochlear fluid, and the micromechanics of the cochlear partition. Most cochlear models are one-dimensional transmission lines with a lumped middle ear[5, 27]. Fluid models in 2D have also been integrated in the time domain[28], and SOAEs and SFOAEs have been computed as the ear-canal pressure[18, 29].

Figure 1. The model. A probe drives the air in the ear canal (PdP_d). The eardrum drives the cochlear fluid through the malleus–incus lever (gg), the incudo-stapedial joint (ISJ), and the stapes. The stapes moves the basal face of the cochlea as a piston. The thin lines indicate the discretization of the fluid in xx and yy. The same micromechanics is attached at each grid point of the BM, and p=0p = 0 at the helicotrema.

Here the cochlea is unrolled into a 2D box. The coordinate xx runs along the partition from the base (x=0x = 0) to the helicotrema (x=Lx = L), and yy is perpendicular to the partition at y=0y = 0. The two scalae, each of height HH, are symmetric about the partition. Only 0≤y≤H0 \le y \le H is therefore modeled, and pp is the pressure difference between the scalae, twice the pressure in the scala vestibuli. The fluid is ideal and incompressible, so that pp satisfies the Laplace equation,

∂2p∂x2+∂2p∂y2=0.(1)\frac{\partial^2 p}{\partial x^2} + \frac{\partial^2 p}{\partial y^2} = 0. \tag*{(1)}

The pressure couples to the displacement ub(x,t)u_b(x,t) of the basilar membrane (BM) at y=0y = 0, positive toward the scala tympani (−y-y), the wall at y=Hy = H is rigid, the stapes, with displacement xsx_s, moves the whole face x=0x = 0 as a piston, and the pressure vanishes at the helicotrema (x=Lx = L). The boundary conditions are thus

∂p∂y=2ρu¨bat y=0,(2)\frac{\partial p}{\partial y}=2\rho\ddot{u}_b \quad\text{at } y=0, \tag*{(2)}
∂p∂y=0at y=H,(3)\frac{\partial p}{\partial y}=0 \quad\text{at } y=H, \tag*{(3)}
∂p∂x=−2ρx¨sat x=0,(4)\frac{\partial p}{\partial x}=-2\rho\ddot{x}_s \quad\text{at } x=0, \tag*{(4)}
p=0at x=L,(5)p=0 \quad\text{at } x=L, \tag*{(5)}

where the dots denote time derivatives. In this study, L=35L = 35 mm, H=7H = 7 mm, and ρ=1000\rho= 1000 kg/m3^3, as in Ref.[30].

Viscous damping of the BM

An ideal fluid carries arbitrarily short waves along the BM. Such waves destabilize active models. The shortest waves of the model, and hence the results of the model, then depend on the grid. In the perilymph, viscosity damps such waves[26]. The damping force on the BM grows with the wavenumber[31]. Following that approach, the pressure is computed for the ideal fluid, and the viscosity enters only as a damping force on the BM.

A damping that grows with the wavenumber is not a local force. Such a damping was earlier written with the local wavenumber of a single traveling wave[31]. In the time domain, the BM carries waves of many wavenumbers at once. The damping is therefore applied to the cosine modes of the BM velocity,

u˙b(x,t)=∑q=0∞v^q(t)cos⁡kqx,kq=(q+12)πL,(6)\dot{u}_b(x,t)=\sum_{q=0}^{\infty}\hat{v}_q(t)\cos k_qx,\qquad k_q=\left(q+\frac{1}{2}\right)\frac{\pi}{L}, \tag*{(6)}

where kqk_q are the wavenumbers that satisfy the conditions at the base and the helicotrema[32].

Each mode receives its own damping force, and FvF_v is their sum,

Fv(x,t)=∑q=0∞8μkqcoth⁡(kqH) v^q(t)cos⁡kqx,(7)F_v(x,t)=\sum_{q=0}^{\infty}8\mu k_q\coth(k_qH)\,\hat{v}_q(t)\cos k_qx, \tag*{(7)}

where μ\mu is the viscosity. The coefficient equates the power of FvF_v to the viscous dissipation of the ideal flow in both scalae[33]. This dissipation is that of the irrotational flow in the bulk of the fluid, not the shear in the rotational viscous layer on the BM[34]. The bulk dissipation dominates for the short waves that the damping must remove[31]. The treatment holds while the viscous layer is thin compared with the wavelength, so that the flow outside the layer stays irrotational. The damping removes the short waves while hardly affecting the long ones. For μ=0.7\mu= 0.7 mPa s, the damping coefficient is 5959 N s/m3^3 for a wavelength of 0.60.6 mm and 350350 N s/m3^3 for 0.10.1 mm, against the BM damping of model 2 of Ref.[30], 880880 N s/m3^3, at x=10x = 10 mm.

Cochlear micromechanics

The partition at each point is the two-degree-of-freedom system of model 2 of Ref.[30]. The two masses of the partition are the BM, with mass m1m_1 and displacement ubu_b, and the tectorial membrane (TM), with mass m2m_2 and displacement utu_t, both positive in the same direction as ubu_b. Their equations of motion are

m1u¨b=p−Fb−Fv,(8)m_1\ddot{u}_b = p - F_b - F_v, \tag*{(8)}
m2u¨t=−Ft,(9)m_2\ddot{u}_t = -F_t, \tag*{(9)}

where FvF_v is the viscous damping of Eq. (7). The forces of the springs, the dampers, and the active element on the BM and the TM are

Fb=c1u˙b+k1ub+c3u˙c+k3uc−γ(c4u˙c+k4uc),(10)F_b = c_1\dot{u}_b + k_1u_b + c_3\dot{u}_c + k_3u_c - \gamma\left(c_4\dot{u}_c + k_4u_c\right), \tag*{(10)}
Ft=c2u˙t+k2ut−c3u˙c−k3uc+γ(c4u˙c+k4uc),(11)F_t = c_2\dot{u}_t + k_2u_t - c_3\dot{u}_c - k_3u_c + \gamma\left(c_4\dot{u}_c + k_4u_c\right), \tag*{(11)}

where uc=ub−utu_c = u_b - u_t. The masses, stiffnesses kik_i, and damping coefficients cic_i are those of model 2 of [30]. The active element is driven by ucu_c, and its force, the active force, is γ(c4u˙c+k4uc)\gamma\left(c_4\dot{u}_c + k_4u_c\right). The active force acts with opposite signs on the BM and the TM. The active force is limited by a cubic term,

γ=γ0(x)(1−uc2usat2),(12)\gamma= \gamma_0(x)\left(1-\frac{u_c^2}{u_{\mathrm{sat}}^2}\right), \tag*{(12)}

where γ0(x)\gamma_0(x) is the gain at low levels and usatu_{\mathrm{sat}} is the saturation displacement, at which the gain falls to zero. This saturation is the only nonlinearity of the model. The active force thus contains the negative damping of a van der Pol oscillator, which decreases as ucu_c approaches usatu_{\mathrm{sat}}. The active force also contains a corresponding cubic stiffness. Cubic saturation has been used in models of SOAEs and distortion products[5, 16].

The cubic term has no linear range and renders the active element passive for ∣uc∣>usat\lvert u_c\rvert> u_{\mathrm{sat}}. This cubic form is therefore valid only for ∣uc∣\lvert u_c\rvert well below usatu_{\mathrm{sat}}, as in the SOAEs of this study (∣uc∣≲7\lvert u_c\rvert\lesssim7 nm, § 6).

In the smooth cochlea, γ0=1\gamma_0 = 1 everywhere. To create a single place of reflection, γ0\gamma_0 is given a step. The gain γ0\gamma_0 is 1 basal to the step and 0.6 apical to the step. An abrupt change of the gain, even a small one, destabilizes a model[19, 20]. For the SOAEs, the step is at 10.5 mm (§ 4 B and V). For the SFOAE, a smaller step, from 1 to 0.98, is placed near the peak of the probe response (§ 4 A). The parameters are those of model 2 of Ref.[30].

Numerical Method

Time integration

As in earlier time-domain solutions[32, 35], the model is discretized in space and integrated in time. The state consists of velocity and displacement of the micromechanics. Because the fluid is incompressible, the pressure is not a state variable. The pressure follows from the state by a single linear solve. The time integration uses the classical fourth-order Runge–Kutta (RK4) method with Δt=5 μs\Delta t = 5\ \mathrm{\mu s}, with the stimulus f(t)f(t), the force on the probe diaphragm (§ 3 C), sampled at intervals of Δt/2\Delta t/2.

The pressure is coupled to the middle ear. The acceleration of the stapes x¨s\ddot{x}_s enters the condition at x=0x = 0, Eq. (4), but is not known when the pressure is computed. The acceleration x¨s\ddot{x}_s depends on the fluid load on the stapes, which in turn depends on the pressure. In the earlier solution[32], the acceleration of the stapes was prescribed, and this loop did not arise. Including the loop directly would put the pressure into the condition at x=0x = 0, whereas the fast solver of § 3 B requires a condition there that is free of the pressure[32]. The loop is therefore closed by superposition. The discretized problem is linear, so the pressure is split as

p=pa+x¨sχ,(13)\mathbf{p} = \mathbf{p}_a + \ddot{x}_s \boldsymbol{\chi}, \tag*{(13)}

where pa\mathbf{p}_a is the solution for x¨s=0\ddot{x}_s = 0, driven by FbF_b and FvF_v alone. The vector χ\boldsymbol{\chi} is the solution for a unit acceleration of the stapes with Fb=Fv=0F_b = F_v = 0. The vector χ\boldsymbol{\chi} does not depend on the state and is computed once in advance.

Each stage consists of three steps, shown in Figure 2.

Figure 2. One stage of the time integration. Step (i) computes the viscous damping FvF_v and the micromechanical forces FbF_b and FtF_t. Step (ii) computes the pressure pa\mathbf{p}_a driven by the BM, the acceleration x¨s\ddot{x}_s of the stapes, and the total pressure p\mathbf{p}. Step (iii) computes the accelerations u¨b\ddot{u}_b and u¨t\ddot{u}_t of the partition.

  • (i) FbF_b and FtF_t are evaluated from the state by Eqs. (10) and (11), and FvF_v by Eq. (7).

  • (ii) The pressure is computed in three parts. First, pa\mathbf{p}_a is solved from FbF_b and FvF_v (§ 3 B). Second, the middle ear gives x¨s\ddot{x}_s from the stimulus f(t)f(t) and from pa\mathbf{p}_a on the face x=0x = 0 (§ 3 C). Third, Eq. (13) gives p\mathbf{p}.

  • (iii) The accelerations of the partition follow from Eqs. (8) and (9). Those of the middle ear follow from § 3 C.

Fluid simulation

Step (ii) of Figure 2 needs the pressure before the BM acceleration is known. Substituting the motion of the BM, Eq. (8), into the condition at the BM, Eq. (2), eliminates the BM acceleration. The result is the Robin condition

∂p∂y−2ρm1p=−2ρm1(Fb+Fv)at y=0.(14)\frac{\partial p}{\partial y} - \frac{2\rho}{m_1}p = -\frac{2\rho}{m_1}(F_b + F_v) \quad\text{at } y = 0. \tag*{(14)}

Step (ii) obtains pa\mathbf{p}_a by solving the Laplace equation, Eq. (1), with this Robin condition, the conditions of Eqs. (3) and (5), and Eq. (4) with x¨s=0\ddot{x}_s = 0, four times per time step. The vector χ\boldsymbol{\chi} of Eq. (13) is the solution with Fb=Fv=0F_b = F_v = 0 and x¨s=1\ddot{x}_s = 1. The viscous damping FvF_v is evaluated at each stage by a pair of discrete cosine transforms. The first gives the mode amplitudes v^q\hat{v}_q of the BM velocity in Eq. (6), and the second the sum of Eq. (7).

The grid is uniform in xx and nonuniform in yy. The steps in yy are fine at the BM, where the short waves near the characteristic place need them (§ 2 B). They grow geometrically toward the wall. The growth is kept mild because the waves reflected at the step and the motion of the oval window give the pressure a part that hardly decays across the scala. A grid that coarsens quickly, as designed for an exponentially decaying pressure[36], loses accuracy near the wall. The second derivative in yy is approximated by the three-point difference for nonuniform steps. The second derivative in xx is approximated by the central difference. The derivative conditions are imposed with ghost nodes, and the node at x=Lx = L is excluded.

The difference operator in yy is independent of xx. The xx direction is therefore diagonalized by the discrete cosine transform[32]. The modes of this transform are the cosine modes cos⁡kqx\cos k_q x of Eq. (6). The discrete problem thus splits into a tridiagonal system in yy for each cosine mode. Each stage needs only two outputs of the fluid (step (ii) of Figure 2), the pressure on the BM, pa(x,0)p_a(x, 0), and the pressure on the face x=0x = 0, pa(0,y)p_a(0, y). The first drives the BM. The second loads the stapes (§ 3 C). The source of each cosine mode lies at y=0y = 0, in the Robin condition. The pressure of a mode is therefore its source times a fixed profile in yy. This profile is computed once in advance for each mode by a direct solution of its tridiagonal system. Each stage then needs only a pair of discrete cosine transforms and a product for each mode. The vector χ\chi is computed once in advance by the same solver.

The cosine transform in xx requires conditions at the ends in xx that do not contain the pressure[32]. The Robin condition at the BM is solved within the tridiagonal system in yy of each cosine mode. If the middle ear were coupled directly, the condition at x=0x = 0 would contain the pressure on the face through the load on the stapes. The condition at x=0x = 0 would become a condition of the Robin type, like Eq. (14), and would couple the cosine modes. The superposition of Eq. (13) keeps the condition at x=0x = 0 free of the pressure.

The grid has Nx=2000N_x = 2000 points in xx and Ny=56N_y = 56 in yy, with a step of 10 µm at the BM. The nodes in xx are xi=iΔxx_i = i \Delta x, i=0,…,Nx−1i = 0, \ldots, N_x - 1, with Δx=L/Nx=17.5\Delta x = L/N_x = 17.5 µm. The condition at x=0x = 0 is imposed by a ghost node, and p=0p = 0 at the excluded node x=Lx = L. The transform is then the discrete cosine transform of type III, and its inverse that of type II. The NyN_y nodes in yy include both walls. The first 20 steps are 10 µm, up to y=0.2y = 0.2 mm. The following 35 steps grow by the factor 1.15, are capped at 0.5 mm, and are scaled together so that the last node lies at y=Hy = H. The steps thus range from 10 µm to 0.47 mm. A 2D model needs a sufficiently fine grid[37]. In preliminary simulations, the time-domain solution was therefore computed on successively refined grids. The estimated errors stayed within the range expected from the order of the difference scheme.

Incorporating the micromechanics and the middle ear

The micromechanics and the viscous damping enter the fluid simulation only through their forces on the BM, FbF_b and FvF_v, in the Robin condition, Eq. (14). These forces follow from the state. The middle ear enters through the acceleration of the stapes x¨s\ddot{x}_s in the condition at x=0x = 0, Eq. (4). Step (ii) obtains x¨s\ddot{x}_s from the middle ear between pa\mathbf{p}_a and p\mathbf{p} (§ 3 A). This section describes that part.

Let AsA_s be the area of the stapes. The fluid load on the stapes is

Aspˉ0=As(pˉa+χˉx¨s),(15)A_s\bar{p}_0 = A_s\left(\bar{p}_a + \bar{\chi}\ddot{x}_s\right), \tag*{(15)}

where the bar denotes the mean over the face x=0x = 0, taken with trapezoidal weights in yy.

The load uses the pressure difference pp itself, without a further factor. The second part of the load is proportional to the acceleration of the stapes. This part acts as an added mass AsχˉA_s\bar{\chi} of the fluid and is moved to the mass of the stapes. Only the known part AspˉaA_s\bar{p}_a then remains as a force on the middle ear.

The middle ear is the lumped model[24] with the parameters of that study. The model of the middle ear includes the probe diaphragm coupled through the air in the ear canal. The three masses of the middle ear are the probe diaphragm, the umbo, and the stapes. Their displacements r=[xd,xm,xs]T\mathbf{r} = [x_d, x_m, x_s]^T are positive toward the cochlea. Their equations of motion are

Mr¨+Cr˙+Kr=f(t)ed−Aspˉaes,(16)\mathbf{M}\ddot{\mathbf{r}} + \mathbf{C}\dot{\mathbf{r}} + \mathbf{K}\mathbf{r} = f(t)\mathbf{e}_d - A_s\bar{p}_a\mathbf{e}_s, \tag*{(16)}

where M\mathbf{M}, C\mathbf{C}, and K\mathbf{K} are the mass, damping, and stiffness matrices of the model[24], and ed=[1,0,0]T\mathbf{e}_d = [1, 0, 0]^T and es=[0,0,1]T\mathbf{e}_s = [0, 0, 1]^T select the probe diaphragm and the stapes. The coupling to the present cochlea changes the model in two ways. The added mass AsχˉA_s\bar{\chi} is added to the mass of the stapes in M\mathbf{M}. The force −Aspˉa-A_s\bar{p}_a acts on the stapes as the load of the cochlea. The ear-canal pressure, which gives the OAE, is

Pd=Kc(Adxd−Amxm),(17)P_d = K_c(A_dx_d - A_mx_m), \tag*{(17)}

where AdA_d and AmA_m are the areas of the probe diaphragm and the eardrum, and KcK_c is the pressure of the air in the ear canal per unit volume displacement. The values are As=6.25 mm2A_s = 6.25\ \mathrm{mm}^2, Ad=75 mm2A_d = 75\ \mathrm{mm}^2, Am=50 mm2A_m = 50\ \mathrm{mm}^2, and Kc=8.5×1010 Pa/m3K_c = 8.5 \times10^{10}\ \mathrm{Pa}/\mathrm{m}^3[24]. The middle ear is integrated in the same RK4 stages as the cochlea.

Emissions from the model

SFOAE suppression tuning

Before the spontaneous emissions, Figure 3 compares the tuning of the SFOAE with that of the BM. The SFOAE-STC is obtained by the suppression method of Ref.[38], with the probe and suppressor frequencies of that study. The SFOAE-STC is the suppressor level at which the residual of PdP_d at the probe frequency fpf_p falls to a criterion. The step of γ0\gamma_0 provides the reflection site that the SFOAE requires[4, 39]. The step is placed at the peak of the probe response, 12.74 mm for the probe of 4 kHz at 30 dB SPL. The STCs depend on the exact position of the step, so those for several positions are averaged. The step, abrupt at one grid point, is placed at 22 positions from 12.18 to 13.02 mm, at every grid point from 12.18 to 12.39 mm and at every fourth from 12.46 to 13.02 mm. The criterion of the SFOAE-STC is a residual of −36 dB SPL. The BM-STC and the BM FTC are computed in the smooth cochlea at the peak of the probe response. The criterion of the BM-STC is a residual of the BM velocity 15 dB below its maximum over the suppressor frequencies, the depth of the BM suppression at the thresholds of the SFOAE-STC. The criterion of the BM FTC is the BM velocity of the probe alone. Each tone is calibrated in PdP_d and turned on with a 5-ms raised-cosine ramp. After 0.1 s, two rectangular windows of 0.1 s are analyzed. The second gives the response, and the first checks that the response is steady. The thresholds

Figure 3. SFOAE suppression tuning curve (SFOAE-STC), BM suppression tuning curve (BM-STC), and BM frequency tuning curve (FTC). The star marks the probe. The thin lines are the SFOAE-STCs for the positions of the step of γ0\gamma_{0}, and the thick line is their mean. The BM curves are those of the smooth cochlea at the peak of the probe response.

are interpolated linearly in level. The slopes are fitted over 0.3 octave below and above the tip, and Q10Q_{10} is computed from the mean SFOAE-STC.

Table I summarizes the shapes of the three curves. The BM FTC has a sharp tip near fpf_p. The slope of the BM FTC above the tip is steeper than that below. This asymmetry is that of the measured BM tuning curves[40].

CurveTipQ10Q_{10}Slope (dB/oct)
(oct)(dB SPL)BelowAbove
SFOAE-STC+0.11831.31.10−1313
BM-STC+0.08431.21.11−1313
BM FTC−0.03730.02.67−2663

Table I. Tip, sharpness, and slopes of the tuning curves of Figure 3. The tip frequency is in octaves re the probe frequency fpf_p. Q10Q_{10} is fpf_p divided by the bandwidth 10 dB above the tip.

The SFOAE-STC and the BM-STC have the same shape. Their tips, sharpnesses, and slopes agree. Both are less than half as sharp as the BM FTC, in particular above fpf_p. The SFOAE suppression thus reflects the suppression of the BM response at the place of the probe rather than the frequency tuning of the BM. The same result was found in measurements, where the SFOAE-STCs were broader than the compound-action-potential STCs[38]. The SFOAE-STC of the model is broader than the measured one, but the trend is the same. A 2D model without a middle ear gave the same result[41].

These results support the model for the emissions that arise by reflection. The BM tuning of the model has the shape found in the measurements. The SFOAE generated by the reflection at the step follows the BM at the place of the reflection, as in the measurements of the SFOAE. The model thus relates the emission in the ear canal to the motion of the BM at the reflection site. The spontaneous emissions below arise from the same kind of reflection.

Spontaneous emissions

Figure 4(a) and Figure 4(b) show the ear-canal pressure PdP_d without stimulus. The emission is a group of equally spaced lines. The main line is denoted by f2f_2. The next strongest line is denoted by f1f_1.

Figure 4. Spontaneous emission for the single step of γ0\gamma_0. The initial BM velocity is uniform, 1 μ\mum/s, and the rest of the state is zero. No stimulus is applied. (a) Waveform of the ear-canal pressure PdP_d band-passed to 2–20 kHz. (b) Spectrum of PdP_d over the last 1 s of a 4-s solution, with a Hann window. The two strongest lines are labeled f2f_2 and f1f_1. (c) Amplitude of the BM velocity at the line f2f_2 in the steady state of the time-domain solution, 11.5–12 s, normalized by its maximum over xx. The dashed and dash-dotted lines are the forward and backward waves of this velocity, shown basal to the step. (d) Phase of the BM velocity at the line f2f_2, zero at the step. (e) Phase differences ψn1n2\psi_{n_1n_2} of Eq. (18) of the intermodulation products relative to their final values, labeled by (n1,n2)(n_1,n_2). The dotted lines in (c) and (d) mark the step.

The emissions of the model share three properties with measured SOAEs. First, each line is narrow-band. Human SOAEs are narrow-band signals whose amplitude and period fluctuate by only a few percent or less[42]. The lines of the model, computed without noise, do not fluctuate at all. Second, the emission arises at the edge of a region of reduced gain. In a guinea pig, an SOAE was accompanied by a spontaneous oscillation of the BM, in an ear whose sensitivity was reduced at frequencies below that of the emission[43]. The model likewise produces the BM oscillation together with the ear-canal pressure. Third, the spectrum contains multiple lines whose phases lock. Human SOAE spectra may contain multiple peaks, and in equally spaced triplets the phases lock as those of a cubic distortion product[13].

The BM oscillation of the main line f2f_2 is a standing wave, as shown in Figure 4(c) and Figure 4(d). These panels show the steady state of the solution. Near the step, the pattern is a traveling wave. On the basal side, the phase is constant. To separate the two waves, the BM velocity vb(x,t)=u˙b(x,t)v_b(x,t)=\dot{u}_b(x,t) is transformed in time and space to vb(k,f)v_b(k,f), where kk is the wavenumber. The transforms use the kernels e−2πifte^{-2\pi i f t} and e−ikxe^{-i k x}, the time window is a Hann window, and the space transform covers the whole length without a window. At the frequency of the main line, the components with k<0k<0 form the forward wave and those with k>0k>0 the backward wave. Each is transformed back to xx. Basal to the step, the standing wave consists of these forward and backward waves, as shown in Figure 4(c). SOAEs have been interpreted as such standing waves[3].

As shown in Figure 4(b), the lines are spaced by Δ=f2−f1\Delta=f_2-f_1. Their frequencies are thus n1f1+n2f2n_1f_1+n_2f_2, with integers n1n_1 and n2n_2. If a line is an intermodulation product of the lines f1f_1 and f2f_2, its phase follows from their phases ϕ1\phi_1 and ϕ2\phi_2 as n1ϕ1+n2ϕ2n_1\phi_1+n_2\phi_2. The actual phase ϕn1n2\phi_{n_1n_2} of each line is compared with this phase by the difference

ψn1n2=ϕn1n2−(n1ϕ1+n2ϕ2).(18)\psi_{n_1n_2} = \phi_{n_1n_2} - (n_1\phi_1 + n_2\phi_2). \tag*{(18)}

This difference stays constant for an intermodulation product and drifts for an independent oscillation. The phases are those of the Hann-windowed Fourier transform of PdP_d at n1f1+n2f2n_1f_1 + n_2f_2, in windows of 0.5 s shifted by 0.1 s. The frequencies f1f_1 and f2f_2 are the peaks of the spectrum over 6–12 s. In the solution, ψn1n2\psi_{n_1n_2} of every line settles to a constant, as shown in Figure 4(e). The lines are thus intermodulation products of the main line f2f_2 and the line f1f_1.

The levels of the emissions are not fitted. Without stimulus, the only nonlinearity is the saturation of Eq. (12). The solution therefore scales with usatu_{\mathrm{sat}}, taken from model 2 of Ref. 30. The frequencies, the spacing, and the stability of the lines do not depend on usatu_{\mathrm{sat}}.

ANALYSIS OF THE EMISSIONS

The main line and the saturation

The analysis starts from the main line f2f_2 of the emission, the strongest line, as shown in Figure 4(b). The other lines are the intermodulation products with the line f1f_1, as shown in Figure 4(e). The oscillation of the main line alone is therefore computed first by harmonic balance, as described in Appendix B. The saturation of Eq. (12) limits the amplitude of the main line.

The harmonic balance calculates uc=Acos⁡ωtu_c = A \cos\omega t at each place, where AA is the amplitude of the main line and ω=2πf2\omega= 2\pi f_2 its angular frequency. The other lines are much weaker, so the main line alone sets the gain. The factor of the gain in Eq. (12) is then

1−uc2usat2=1⏟linear−A22usat2⏟cross-saturation−A22usat2cos⁡2ωt⏟pump.(19)1-\frac{u_c^2}{u_{\mathrm{sat}}^2}=\underbrace{1}_{\text{linear}}-\underbrace{\frac{A^2}{2u_{\mathrm{sat}}^2}}_{\text{cross-saturation}}-\underbrace{\frac{A^2}{2u_{\mathrm{sat}}^2}\cos2\omega t}_{\text{pump}}. \tag*{(19)}

Equation (19) shows the factor of the gain alone. The active force is this factor times c4u˙c+k4ucc_4\dot{u}_c+k_4u_c. The exact linearization in Appendix B differentiates this product, so the damping and the stiffness are lowered by different amounts.

The three terms correspond to the three kinds of eigenvalues in § 6 C and V D. The first term gives the linear model, analyzed in § 6 C. The second term is always negative.

The second term thus lowers the mean gain where the main line is large. Every other line sees this lower gain. This lowering of the mean gain is the cross-saturation.

The pump

The third term of Eq. (19) oscillates at 2ω2\omega. The third term thus modulates the gain in time. The factor cos⁡2ωt\cos2\omega t is this oscillation of the gain. Consider a line shifted by Ω\Omega from the main line, cos⁡[(ω+Ω)t]\cos[(\omega+\Omega)t]. The active force on this line is the gain times the line. The oscillation of the gain multiplies the line as

cos⁡2ωtcos⁡[(ω+Ω)t]=12cos⁡[(ω−Ω)t]+12cos⁡[(3ω+Ω)t].\cos2\omega t\cos[(\omega+\Omega)t]=\frac{1}{2}\cos[(\omega-\Omega)t]+\frac{1}{2}\cos[(3\omega+\Omega)t].

The component at 3ω+Ω3\omega+\Omega is far from the lines and is neglected. The other component is at the difference frequency 2ω−(ω+Ω)=ω−Ω2\omega-(\omega+\Omega)=\omega-\Omega. The line at ω+Ω\omega+\Omega thus drives a line at ω−Ω\omega-\Omega.

In the same way, a line at ω−Ω\omega-\Omega drives one at ω+Ω\omega+\Omega. The oscillation of the gain thus couples two lines placed symmetrically about the main line, as shown in Figure 5(a). This oscillating

Figure 5. (a) The pump at 2ω2\omega and the two lines at ω±Ω\omega\pm\Omega. Each line drives its mirror image about the main line. (b) The line at ω−Ω\omega- \Omega (solid), the component driven by the line at ω+Ω\omega+ \Omega through the pump (dashed), and the sum of the two (thick). The component is in phase with the line (top) and in opposite phase (bottom). (c) The same at ω+Ω\omega+ \Omega, with the component driven by the line at ω−Ω\omega- \Omega.

gain is the pump introduced in § 1. The coupling is the four-wave mixing of a saturable gain[22].

Whether the drive by the pump raises or lowers a line depends on the phase. The line at ω±Ω\omega\pm\Omega drives a component at ω∓Ω\omega\mp\Omega. If each component is in phase with the line at the same frequency, the sum is larger than the line alone, as shown in the top of Figure 5(b) and Figure 5(c). The component then amplifies the line. If the component is in opposite phase, the sum is smaller, as shown in the bottom of Figure 5(b) and Figure 5(c). The component then suppresses the line. An amplified line drives the other line more strongly. Through this loop, shown in Figure 5(a), the two lines amplify each other in phase and suppress each other in opposite phase. If this mutual amplification exceeds the decay of each line alone, the pair grows.

The pump acts on a pair at any Ω\Omega. The decay of each line alone, however, depends on the frequency of the line. A line decays slowly only near a weakly damped mode of the cochlea. A pair can therefore grow only at a particular Ω\Omega, where one of the two lines lies near such a mode. In the model, the weakly damped mode lies just above the line f1f_1.

Linear eigenvalues

Without stimulus (f=0f = 0), the state of rest is an equilibrium of the model. The stability of this state decides whether an SOAE arises, and at which frequencies. As in state-space models[16, 19, 20], the stability is analyzed by the eigenvalues of the model linearized about rest. Here the model includes a 2D fluid and the middle ear. Since the bracket of Eq. (12) differs from 1 only at second order in ucu_c, the linearized model is the model with γ=γ0(x)\gamma= \gamma_0(x).

The discretized model is a system of first-order equations w˙=F(w,f)\dot{\mathbf{w}} = \mathbf{F}(\mathbf{w}, f) for a state vector w\mathbf{w} that contains the displacements and velocities of the BM and the TM at the NxN_x grid points. The vector also contains those of the probe diaphragm, the umbo, and the stapes. The model thus has 4Nx+64N_x + 6 state variables. Without stimulus and linearized about rest, the system is

w˙=Jw.(20)\dot{\mathbf{w}} = \mathbf{J}\mathbf{w}. \tag*{(20)}

The matrix J\mathbf{J} is the Jacobian matrix of F\mathbf{F} at rest and is the state matrix of the linearized model[16, 19, 20]. Appendix A describes the computation of the Jacobian J\mathbf{J}.

The eigenvalues sm=σm+iωms_m = \sigma_m + i\omega_m of J\mathbf{J} give the modes of the model. For ωm>0\omega_m > 0, ωm/2π\omega_m/2\pi is the frequency of mode mm and σm\sigma_m its growth rate. If all σm<0\sigma_m < 0, the state of rest is stable and the model has no SOAE. If any σm>0\sigma_m > 0, the state of rest is unstable. The cubic term of Eq. (12) then limits the oscillation. The amplitude of the oscillation is found in the time domain.

The linear eigenvalues of J\mathbf{J} show several unstable modes, as the triangles in Figure 6(a) show. The unstable mode nearest to the main line f2f_2 is called mode 2, and that nearest to the line f1f_1 is called mode 1.

Figure 6. Linear eigenvalues of JJ (triangles) and quasilinear eigenvalues of Eq. (23) with the cross-saturation added (squares) and with the pump also added (circles). Filled markers are unstable, and open markers stable. Panel (b) magnifies the pair and the main line of (a). Each eigenvalue with the pump consists of two lines placed symmetrically about the main line and is plotted at both. The pair closest to zero is unstable. The circle at the main line with s=0s = 0 is the main line itself. The dotted lines mark the main line and the two lines of the pair.

Quasilinear eigenvalues

The linear eigenvalues of J\mathbf{J} do not include the cross-saturation and the pump by the main line. The model is linearized about the oscillation of the main line. The pair is small compared with the main line, so the linearization holds. The linearized model follows from Eq. (20) with the gain factor of Eq. (19).

The first term of Eq. (19) gives J\mathbf{J}. The second term is constant in time. This constant lowering of the gain turns J\mathbf{J} into another fixed matrix J0\mathbf{J}_0, as described in Appendix B. The third term, in contrast, varies in time as cos⁡2ωt\cos2\omega t. This term cannot be put into a fixed matrix in the same way. The third term is therefore kept as a separate matrix multiplied by the time factor. By Euler’s formula, cos⁡2ωt=(e2iωt+e−2iωt)/2\cos2\omega t = (e^{2i\omega t} + e^{-2i\omega t})/2. The third term thus gives a matrix P\mathbf{P} multiplied by e2iωte^{2i\omega t} and the conjugate matrix P‾\overline{\mathbf{P}} multiplied by e−2iωte^{-2i\omega t}. The gain is real, so the two matrices P\mathbf{P} and P‾\overline{\mathbf{P}} are conjugate to each other. Appendix B gives J0\mathbf{J}_0 and P\mathbf{P} from the linearization of the cubic term, in Eqs. (31) and (B5).

The linearized model is thus

w˙=J0w+(Pe2iωt+P‾e−2iωt)w.(21)\dot{\mathbf{w}} = \mathbf{J}_0 \mathbf{w} + \left(\mathbf{P}e^{2i\omega t} + \overline{\mathbf{P}}e^{-2i\omega t}\right)\mathbf{w}. \tag*{(21)}

The second term is the pump. The pump couples only the two lines at ω±Ω\omega\pm\Omega, as described in § 6 B. Solutions are sought as two lines that grow together at the rate σ\sigma. The line at ω+Ω\omega+ \Omega is then eσtei(ω+Ω)t=e(s+iω)t≡e+e^{\sigma t}e^{i(\omega+\Omega)t} = e^{(s+i\omega)t} \equiv e_+, with s=σ+iΩs = \sigma+ i\Omega. The conjugate of the line at ω−Ω\omega- \Omega is eσte−i(ω−Ω)t=e(s−iω)t≡e−e^{\sigma t}e^{-i(\omega-\Omega)t} = e^{(s-i\omega)t} \equiv e_-, with the same ss. Equation (21) is linear and has real coefficients. The real part of a complex solution is thus also a solution. The state is therefore sought as the complex solution

w=c+e++c−e−.(22)\mathbf{w} = \mathbf{c}_+e_+ + \mathbf{c}_-e_-. \tag*{(22)}

Here c+\mathbf{c}_+ and c−\mathbf{c}_- are constant complex vectors over the state. Equation (22) is substituted into Eq. (21). The time derivative and the pump act on the two terms of Eq. (22) as

ddt(c±e±)=(s±iω)c±e±,\frac{d}{dt}\left(\mathbf{c}_{\pm}e_{\pm}\right) = (s \pm i\omega)\mathbf{c}_{\pm}e_{\pm},
e±2iωte∓=e±.e^{\pm2i\omega t}e_{\mp} = e_{\pm}.

The pump thus turns each line into the other. The other products of the pump oscillate near 3ω3\omega and are neglected. The factors e+e_+ and e−e_- oscillate at different frequencies. The terms with each factor must therefore agree on the two sides of Eq. (21),

(s+iω)c+=J0c++Pc−,(s+i\omega)\mathbf{c}_{+}=\mathbf{J}_{0}\mathbf{c}_{+}+\mathbf{P}\mathbf{c}_{-},
(s−iω)c−=J0c−+P‾c+.(s-i\omega)\mathbf{c}_{-}=\mathbf{J}_{0}\mathbf{c}_{-}+\overline{\mathbf{P}}\mathbf{c}_{+}.

Moving ±iωc±\pm i\omega\mathbf{c}_{\pm} to the right sides gives

s(c+c−)=(J0−iωPP‾J0+iω)(c+c−).(23)s\begin{pmatrix}\mathbf{c}_{+}\\ \mathbf{c}_{-}\end{pmatrix} = \begin{pmatrix} \mathbf{J}_{0}-i\omega& \mathbf{P}\\ \overline{\mathbf{P}} & \mathbf{J}_{0}+i\omega \end{pmatrix} \begin{pmatrix}\mathbf{c}_{+}\\ \mathbf{c}_{-}\end{pmatrix}. \tag*{(23)}

The oscillating factors have disappeared, and the coefficients are fixed matrices. The eigenvalues ss of Eq. (23) are called the quasilinear eigenvalues. Each eigenvector consists of two lines placed symmetrically about the main line. Without the pump (P = 0), the two lines decouple. The quasilinear eigenvalues are then those of J0\mathbf{J}_{0}, measured from the main line.

One eigenvalue with the pump is s=0s=0. This eigenvalue is the main line itself, shifted in time.

Without the pump, all the modes decay, as the squares in Figure 6 show. Mode 1 also decays and lies above the line f1f_{1}. The mode of the main line decays too, because J0\mathbf{J}_{0} applies the full cross-saturation to this mode. The main line itself sees only half of this lowering, the self-saturation (Appendix B). Without the pump, the saturation thus stabilizes all the modes other than the main line. The saturation alone favors a single line.

The pump destabilizes one pair of lines, as shown in Figure 6(a) and Figure 6(b). The pair lies at the line f1f_{1} and at the mirror image of f1f_{1} about the main line, the line f2+Δf_{2}+\Delta. All the other pairs decay. Without the pump, each line of the pair decays. The growth of the pair is thus the effect of the pump. The pump also equalizes the spacing of the lines.

The time-domain solution confirms this pair. The model is solved from an initial condition that contains only the mode of the main line. This mode is the eigenvector nearest to f2f_2 of the model with the self-saturation of Eq. (29), scaled to the peak BM velocity of the main line. The initial condition is not the harmonic-balance solution, so the main line first adjusts to its steady state. Two sidebands at f2±Δf_2 \pm\Delta then grow together out of the main line, as Figure 7(a) shows. The single main line turns into a group of lines, as shown in Figure 7(b). The lower sideband lies at the frequency of the line f1f_1. The growth rate of the sidebands is close to that of the pair. The oscillation of the main line alone is therefore unstable to modulation. The other lines of the emission are phase-locked intermodulation products of the main line and the line f1f_1 (§ 4 B). The line f1f_1 itself thus arises from the main line alone.

Figure 7. Time-domain solution started from the mode of the main line alone. (a) Levels of the main line f2f_2 and of the sidebands f2±Δf_2 \pm\Delta in the ear-canal pressure PdP_d, from a 40-ms Hann window shifted by 10 ms. (b) Spectrum of PdP_d over 0.6–1.0 s.

Discussion

The pump creates the lines

The pump destabilizes one pair of lines, and this pair produces the group of lines. The unstable pair is shown by the filled circles in Figure 6(b). The pair lies at the line f1f_1 and at the line f2+Δf_2 + \Delta. The other lines of the emission are phase-locked intermodulation products of the main line and the pair, as Figure 4(e) shows.

Without the pump, the cross-saturation stabilizes all the modes, as the squares in Figure 6(a) show. Several modes are unstable in the linear eigenvalues, shown by the triangles, but all of these modes decay with the cross-saturation. That not all unstable modes become SOAEs was found in state-space models[16]. The cross-saturation gives a reason for this finding.

A pump generated by the main line itself has not been stated for a cochlear model, but similar mechanisms are known in other oscillators. In earlier measurements, an external low-frequency tone modulated the amplitude of SOAEs and produced sidebands[23]. In the model, the main line itself modulates the gain of the active element. In delayed-feedback oscillators, modes that are damped alone are likewise excited on the background of the main oscillation[44]. In optical microresonators, four-wave mixing equalizes the spacing of the sidebands against the dispersion, by a nonlinear refractive index[45]. In quantum cascade lasers with a fast saturable gain, four-wave mixing locks the modes into a comb[46, 47].

The two explanations of multiple SOAEs

In the model, the two explanations of multiple SOAEs hold together. The standing-wave view accounts for the origin of the main line and for the spacing[3, 10]. The main line is a standing wave between the step and the stapes, as the forward and backward waves in Figure 4(c) show. The pair grows near the neighboring linear modes, as Figure 6(b) shows, and the pump equalizes the spacing of the lines in Figure 4(b). The intermodulation view accounts for the nature of the other lines. The phases of the other lines lock as those of intermodulation products, as Figure 4(e) shows. The phases of equally spaced triplets of human SOAEs lock in the same way[13].

The lines grow from one oscillation, not from independent oscillations. Locked phases alone do not show a single oscillation, since coupled independent oscillators also lock[9, 17]. Without the pump, all the modes decay, as Figure 6(a) shows. The time-domain solution started from the main line alone produces the sidebands, as Figure 7 shows. In the model, the lines are therefore neither independent resonances nor intermodulation products of independent oscillations[16].

The mechanism concerns a group of lines from one standing wave, not SOAE spectra in general. Most neighboring human SOAEs are not related by intermodulation[14]. In barn owls, by contrast, consecutive SOAEs are often nearly equally spaced. This near-equal spacing has been attributed either to intermodulation or to a characteristic delay, such as that of a standing wave[14]. The mechanism of the model would allow both at once. The avian basilar papilla, however, differs from the mammalian cochlea of the model, with a thick tectorial plate and many hair cells across its width. Whether the papilla supports such a standing wave is not addressed by the model.

Standing wave or oscillator

Spectra with several lines, raised in § 1 as a test of the two views of SOAEs, do not by themselves decide between the two views. In the model, the whole standing wave acts as one self-sustained oscillation, and the lines come from this one oscillation, as shown in § 7 B. In this respect, the model is close to the oscillator view. The neighboring resonances that set the spacing, however, come from the standing wave. Whether a local oscillator without such resonances could produce the same lines is not known. The model thus does not exclude other views of SOAEs, such as coupled Hopf oscillators[7, 8] or fluid-coupled elements without a regular reflection site[2].

Assumptions and limitations of the model

The distribution of the gain of the active element in the model differs from that in real ears. The model has a single step in the gain, as at the edge of a region of hearing loss[43].

In real ears, distributed roughness is the prevailing explanation of the reflection[3, 4], and a single site has been judged insufficient to account for spontaneous BM oscillations[48]. In the model, the distribution of the gain can be set freely, and the step places the reflection at a known position. A single step yields a group of lines from one place, not the many SOAEs of a real ear[49, 50]. With distributed roughness, the resonances would not be equally spaced. The pump still couples only lines placed symmetrically about the main line, and a pair grows when one of its lines lies near a weakly damped mode (§ 6 B). The mirror line need not lie on a resonance. Whether the pair still grows when roughness shifts and damps the neighboring modes irregularly was not examined.

The round-trip delay of the model is longer than in human ears. The delay computed from the line spacing Δ\Delta is about twice the OAE group delay of human ears at high frequencies[11]. A shorter delay would double the spacing of the resonances and move the neighboring modes farther from the main line. Whether the pump then still overcomes the cross-saturation was not examined.

The saturation is simplified in two ways. The active element saturates as a cubic and generates no even harmonics[51]. An asymmetric saturation, such as that of the transduction of the outer hair cells, would also modulate the gain at ω\omega and shift its mean. The modulation at ω\omega couples a line at ω+Ω\omega+ \Omega to one at Ω\Omega, far from the main line, whereas the pump at 2ω2\omega couples the pair about the main line. How the asymmetry changes the balance between the cross-saturation and the pump was not examined. The main line is treated by a describing function, which holds because a single line dominates[16]. The quasilinear eigenvalues describe the onset of the pair. In the steady state, the line f1f_1 is no longer small compared with the main line, and the pair then also saturates the main line. The steady state is therefore given by the time-domain solution, not by the eigenvalues.

CONCLUSIONS

A group of equally spaced SOAE lines can arise from a single cochlear standing wave. The cross-saturation by the main line suppresses the other lines. The pump destabilizes a pair of lines near the neighboring resonances. The other lines are intermodulation products of the main line and the pair. In this mechanism, the standing-wave view and the intermodulation view hold together.

ACKNOWLEDGMENTS

This work was supported by JSPS KAKENHI Grant Number JP26K14881. A generative AI tool, Claude (Anthropic), was used to assist with language editing and computer programming. The author reviewed and edited all output and takes full responsibility for the content of this article.

AUTHOR DECLARATIONS

Conflict of Interest

The author has no conflicts to disclose.

Data Availability

The data that support the findings of this study are available from the author upon reasonable request.

State Matrix

This appendix describes how the state matrix J\mathbf{J} of Eq. (20) is computed. The linearization is made on the discretized model of Sec. IIIA, written as a system of first-order equations for the state vector

w=[ubT,u˙bT,utT,u˙tT,rT,r˙T]T(24)\mathbf{w} = [\mathbf{u}_b^{T}, \dot{\mathbf{u}}_b^{T}, \mathbf{u}_t^{T}, \dot{\mathbf{u}}_t^{T}, \mathbf{r}^{T}, \dot{\mathbf{r}}^{T}]^{T} \tag*{(24)}

of 4Nx+64N_x + 6 components. The vectors ub\mathbf{u}_b and ut\mathbf{u}_t contain the displacements at the grid points of the partition. The vector r\mathbf{r} contains the displacements of the probe diaphragm, the umbo, and the stapes of Sec. IIIC. In this state, the discretized model is the system of ordinary differential equations integrated in time (Sec. IIIA),

w˙=F(w,f),(25)\dot{\mathbf{w}} = \mathbf{F}(\mathbf{w}, f), \tag*{(25)}

where F\mathbf{F} is the time derivative of the state that one stage of the integration returns. The vector F\mathbf{F} consists of u˙b\dot{\mathbf{u}}_b, u˙t\dot{\mathbf{u}}_t, and r˙\dot{\mathbf{r}}, and the accelerations given by steps (i)–(iii) of Figure 2. The fluid has no state of its own. Because the fluid is incompressible, the fluid pressure is fixed at each instant by w\mathbf{w} through the linear solve of step (ii) of Figure 2 (Sec. IIIC). The pressure is thus computed within F\mathbf{F} but is not part of w\mathbf{w}.

With γ=γ0(x)\gamma= \gamma_0(x) and f=0f = 0, F(w)≡F(w,0)\mathbf{F}(\mathbf{w}) \equiv\mathbf{F}(\mathbf{w}, 0) is linear in w\mathbf{w} and equals Jw\mathbf{J}\mathbf{w} of Eq. (20). The matrix J\mathbf{J} has (4Nx+6)×(4Nx+6)(4N_x + 6) \times(4N_x + 6) entries.

The entries of J\mathbf{J} are not derived by hand but computed from the discretized equations of motion. Let ej\mathbf{e}_j be the state in which only the jjth component is 1. Any state is w=∑jwjej\mathbf{w} = \sum_j w_j\mathbf{e}_j, and the linearity of F\mathbf{F} gives

F(w)=∑j=1NwjF(ej),N=4Nx+6,(26)\mathbf{F}(\mathbf{w}) = \sum_{j=1}^{N} w_j\mathbf{F}(\mathbf{e}_j), \qquad N = 4N_x + 6, \tag*{(26)}

so that the matrix

J=[F(e1), F(e2), …, F(eN)](27)\mathbf{J} = [\mathbf{F}(\mathbf{e}_1),\ \mathbf{F}(\mathbf{e}_2),\ \ldots,\ \mathbf{F}(\mathbf{e}_N)] \tag*{(27)}

satisfies Eq. (20) for every w\mathbf{w}. Each column is computed by one evaluation of the right-hand side, that is, by one stage of the time integration without advancing in time. This stage is shown in Figure 2. For random w\mathbf{w}, Jw\mathbf{J}\mathbf{w} reproduces F(w)\mathbf{F}(\mathbf{w}) to machine precision. All eigenvalues and eigenvectors of J\mathbf{J} (8006×80068006 \times8006 for Nx=2000N_x = 2000) are computed by the QR algorithm (LAPACK dgeev).

The main line and the pump

This appendix gives the oscillation of the main line and J0J_0 and P\mathbf{P} of Eq. (23). The model is written as

w˙=Jw+BN,N=−γ0usat2uc2(c4u˙c+k4uc),(28)\dot{\mathbf{w}} = \mathbf{J}\mathbf{w} + \mathbf{B}\mathcal{N}, \qquad\mathcal{N} = -\frac{\gamma_0}{u_{\mathrm{sat}}^2}u_c^2(c_4\dot{u}_c + k_4u_c), \tag*{(28)}

where N\mathcal{N} is the cubic part of the active force at each grid point and B\mathbf{B} maps N\mathcal{N} to w˙\dot{\mathbf{w}}.

On a single line, uc=Acos⁡ωtu_c = A\cos\omega t, the cubic term also produces the frequency 3ω3\omega, but only the component of the cubic term at ω\omega affects the line. With cos⁡3ωt=34cos⁡ωt+14cos⁡3ωt\cos^3\omega t = \frac{3}{4}\cos\omega t + \frac{1}{4}\cos3\omega t and cos⁡2ωtsin⁡ωt=14sin⁡ωt+14sin⁡3ωt\cos^2\omega t\sin\omega t = \frac{1}{4}\sin\omega t + \frac{1}{4}\sin3\omega t, this component is

Nω=−γ0S(c4u˙c+3k4uc),S=A24usat2.(29)\mathcal{N}_{\omega} = -\gamma_0S(c_4\dot{u}_c + 3k_4u_c), \qquad S = \frac{A^2}{4u_{\mathrm{sat}}^2}. \tag*{(29)}

The line thus sees the gains γ0(1−S)\gamma_0(1-S) in the damping and γ0(1−3S)\gamma_0(1-3S) in the stiffness. This reduction is the self-saturation.

The oscillation of the main line, w=Re⁡(Weiωt)\mathbf{w}=\operatorname{Re}(\mathbf{W}e^{i\omega t}), is found from iωW=JW+BNωi\omega\mathbf{W}=\mathbf{J}\mathbf{W}+\mathbf{B}\mathcal{N}_{\omega}, where Nω\mathcal{N}_{\omega} is the component of N\mathcal{N} at ω\omega of Eq. (29). As a phasor, Nω\mathcal{N}_{\omega} has the entries −γ0S(iωc4+3k4)U-\gamma_0S(i\omega c_4+3k_4)U, with UU the entries of CuWC_u\mathbf{W} of the next paragraph and S=∣U∣2/4usat2S=\lvert U\rvert^2/4u_{\mathrm{sat}}^2.

Since SS is not analytic in UU, the equation is split into its real and imaginary parts. The solution is fixed in phase by setting the imaginary part of u˙b\dot{u}_b to zero at the node where ∣u˙b∣\lvert\dot{u}_b\rvert of the initial guess is largest. The resulting 2N+12N+1 real equations are solved for the real and imaginary parts of W\mathbf{W} and for ω\omega by Newton’s method, starting from an eigenvector of the mode of the main line.

Let the main line be uc=Re⁡(Ueiωt)u_c=\operatorname{Re}(Ue^{i\omega t}) at each grid point, and let S=U2/4usat2S=U^2/4u_{\mathrm{sat}}^2, so that ∣S∣=S\lvert S\rvert=S. A small change of ucu_c and u˙c\dot{u}_c changes N\mathcal{N} by the derivatives ∂N/∂uc\partial\mathcal{N}/\partial u_c and ∂N/∂u˙c\partial\mathcal{N}/\partial\dot{u}_c, taken on the oscillation of the main line. With u˙c=Re⁡(iωUeiωt)\dot{u}_c=\operatorname{Re}(i\omega Ue^{i\omega t}), these derivatives are

∂N∂u˙c=−γ0c4(2S+Se2iωt+Sˉe−2iωt),∂N∂uc=−γ0[6Sk4+(3k4+2iωc4)Se2iωt+(3k4−2iωc4)Sˉe−2iωt].(30)\begin{aligned} \frac{\partial\mathcal{N}}{\partial\dot{u}_c} &= -\gamma_0c_4\left(2S+Se^{2i\omega t}+\bar{S}e^{-2i\omega t}\right),\\ \frac{\partial\mathcal{N}}{\partial u_c} &= -\gamma_0\left[6Sk_4+(3k_4+2i\omega c_4)Se^{2i\omega t}+(3k_4-2i\omega c_4)\bar{S}e^{-2i\omega t}\right]. \tag*{(30)} \end{aligned}

Let CuC_u and CvC_v pick ucu_c and u˙c\dot{u}_c at the grid points from w\mathbf{w}, and let SS and S\mathcal{S} also denote the diagonal matrices of their values at the grid points. The matrix S\mathcal{S} is real, and S\mathcal{S} is complex, with the conjugate Sˉ\bar{\mathcal{S}}. The constant parts give

J0=J−γ0B(6Sk4Cu+2Sc4Cv).(31)J_0=J-\gamma_0B\left(6Sk_4C_u+2Sc_4C_v\right). \tag*{(31)}

The matrix J0J_0 thus has the gains γ0(1−2S)\gamma_0(1-2S) in the damping and γ0(1−6S)\gamma_0(1-6S) in the stiffness. This lowering is the cross-saturation. The cross-saturation is thus twice the self-saturation.

of Eq. (29). The parts at e2iωte^{2i\omega t} give the pump

P=−γ0B[(3k4+2iωc4)SCu+c4SCv].(32)\mathrm{P}=-\gamma_{0}\mathrm{B}\left[(3k_{4}+2i\omega c_{4})\mathrm{SC}_{u}+c_{4}\mathrm{SC}_{v}\right]. \tag*{(32)}

The parts at e−2iωte^{-2i\omega t} give P‾\overline{\mathrm{P}}. These parts give Eq. (21). Substituting the complex solution

c+e(s+iω)t+c−e(s−iω)tc_{+}e^{(s+i\omega)t}+c_{-}e^{(s-i\omega)t}

into Eq. (21) and collecting the terms at e(s+iω)te^{(s+i\omega)t} and e(s−iω)te^{(s-i\omega)t} gives Eq. (23). Terms at ±3ω\pm3\omega are neglected.

The eigenvalues of Eq. (23), a matrix of size 2(4Nx+6)2(4N_{x}+6), are found by shift-and-invert with shifts every 200 Hz, from 5.4 to 7.2 kHz. Since ss and s‾\overline{s} describe the same pair with the two lines exchanged, the shifts cover both sides of the main line. These eigenvalues are the Floquet exponents of the oscillation of the main line, truncated at ±ω\pm\omega.

References

  1. [1]1D. T. Kemp, “Stimulated acoustic emissions from within the human auditory system,” The Journal of the Acoustical Society of America 64(5), 1386–1391 (1978) doi: 10.1121/1.382104.DOI
  2. [2]2H. P. Wit, “A time-domain model for the generation of spontaneous otoacoustic emissions in resonant cavities,” Journal of the Association for Research in Otolaryngology 27(5), 785–804 (2026) doi: 10.1007/s10162-026-01063-x.DOI
  3. [3]3C. A. Shera, “Mammalian spontaneous otoacoustic emissions are amplitude-stabilized cochlear standing waves,” J. Acoust. Soc. Am. 114, 244–262 (2003).pubmed.ncbi.nlm.nih.gov/12880039
  4. [4]4G. Zweig and C. A. Shera, “The origin of periodicity in the spectrum of evoked otoacoustic emissions,” J. Acoust. Soc. Am. 98(4), 2018–2047 (1995) doi: 10.1121/1.413320.DOI
  5. [5]C. L. Talmadge, A. Tubis, G. R. Long, and P. Piskorski, “Modeling otoacoustic emission and hearing threshold fine structures,” J. Acoust. Soc. Am. 104(3), 1517–1543 (1998).pubmed.ncbi.nlm.nih.gov/9745736
  6. [6]S. Camalet, T. Duke, F. Jülicher, and J. Prost, “Auditory sensitivity provided by self-tuned critical oscillations of hair cells,” Proceedings of the National Academy of Sciences 97(7), 3183–3188 (2000) doi: 10.1073/pnas.97.7.3183.DOI
  7. [7]A. J. Hudspeth, F. Jülicher, and P. Martin, “A critique of the critical cochlea: Hopf—a bifurcation—is better than none,” Journal of Neurophysiology 104(3), 1219–1229 (2010) doi: 10.1152/jn.00437.2010.DOI
  8. [8]F. Fruth, F. Jülicher, and B. Lindner, “An active oscillator model describes the statistics of spontaneous otoacoustic emissions,” Biophysical Journal 107(4), 815–824 (2014) doi: 10.1016/j.bpj.2014.06.047.DOI
  9. [9]H. P. Wit and P. van Dijk, “Are human spontaneous otoacoustic emissions generated by a chain of coupled nonlinear oscillators?,” The Journal of the Acoustical Society of America 132(2), 918–926 (2012) doi: 10.1121/1.4730886.DOI
  10. [10]C. A. Shera, “Whistling while it works: Spontaneous otoacoustic emissions and the cochlear amplifier,” J. Assoc. Res. Otolaryngol. 23, 17–25 (2022) doi: 10.1007/s10162–021–00829–9.DOI
  11. [11]C. A. Shera and J. J. Guinan, “Stimulus-frequency-emission group delay: A test of coherent reflection filtering and a window on cochlear tuning,” The Journal of the Acoustical Society of America 113(5), 2762–2772 (2003) doi: 10.1121/1.1557211.DOI
  12. [12]J. B. Dewey, “Sources of microstructure in mammalian cochlear responses,” Journal of the Association for Research in Otolaryngology 26(1), 1–15 (2025) doi: 10.1007/s10162-025-00974-5.DOI
  13. [13]P. van Dijk and H. P. Wit, “Synchronization of cubic distortion spontaneous otoacoustic emissions,” J. Acoust. Soc. Am. 104(1), 591–594 (1998) doi: 10.1121/1.423287.DOI
  14. [14]C. Bergevin, R. E. Whiley, H. Wit, G. A. Manley, and P. van Dijk, “Auditory cellular cooperativity probed via spontaneous otoacoustic emissions,” Biophysical Journal 124(8), 1208–1225 (2025) doi: 10.1016/j.bpj.2025.02.023.DOI
  15. [15]W. S. Rhode, “Mutual suppression in the 6 kHz region of sensitive chinchilla cochleae,” The Journal of the Acoustical Society of America 121(5), 2805–2818 (2007) doi: 10.1121/1.2718398.DOI
  16. [16]E. M. Ku, S. J. Elliott, and B. Lineton, “Limit cycle oscillations in a nonlinear state space model of the human cochlea,” J. Acoust. Soc. Am. 126(2), 739–750 (2009) doi: 10.1121/1.3158861.DOI
  17. [17]M. Gelfand, O. Piro, M. O. Magnasco, and A. J. Hudspeth, “Interactions between hair cells shape spontaneous otoacoustic emissions in a model of the tokay gecko’s cochlea,” PLoS ONE 5(6), e11116 (2010) doi: 10.1371/journal.pone.0011116.DOI
  18. [18]J. Meaud, T. Bowling, and C. Lemons, “Computational modeling of spontaneous otoacoustic emissions by the mammalian cochlea,” in Proceedings of the ASME 2018 Dynamic Systems and Control Conference (2018), Vol. 1, p. V001T11A001, doi: 10.1115/DSCC2018-9044.
  19. [19]19S. J. Elliott, E. M. Ku, and B. Lineton, “A state space model for cochlear mechanics,” J. Acoust. Soc. Am. 122(5), 2759–2771 (2007) doi: 10.1121/1.2783125.DOI
  20. [20]20E. M. Ku, S. J. Elliott, and B. Lineton, “Statistics of instabilities in a state space model of the human cochlea,” J. Acoust. Soc. Am. 124(2), 1068–1079 (2008) doi: 10.1121/1.2939133.DOI
  21. [21]21W. E. Lamb, Jr., “Theory of an optical maser,” Phys. Rev. 134, A1429–A1450 (1964) doi: 10.1103/PhysRev.134.A1429.DOI
  22. [22]22G. P. Agrawal, “Population pulsations and nondegenerate four-wave mixing in semiconductor lasers and amplifiers,” J. Opt. Soc. Am. B 5, 147–159 (1988) doi: 10.1364/JOSAB.5.000147.DOI
  23. [23]23L. Bian and K. L. Watts, “Effects of low-frequency biasing on spontaneous otoacoustic emissions: Amplitude modulation,” The Journal of the Acoustical Society of America 123(2), 887–898 (2008) doi: 10.1121/1.2821983.DOI
  24. [24]24Y.-W. Liu and S. T. Neely, “Distortion product emissions from a cochlear model with nonlinear mechanoelectrical transduction in outer hair cells,” J. Acoust. Soc. Am. 127(4), 2420–2432 (2010).DOI
  25. [25]25T. Bowling, C. Lemons, and J. Meaud, “Reducing tectorial membrane viscoelasticity enhances spontaneous otoacoustic emissions and compromises the detection of low level sound,” Sci. Rep. 9, 7494 (2019) doi: 10.1038/s41598-019-43970-5.DOI
  26. [26]26R. Nobili, A. Vetešník, L. Turicchia, and F. Mammanao, “Otoacoustic emissions from residual oscillations of the cochlear basilar membrane in a human ear model,” J. Assoc. Res. Otolaryngol. 4, 478–494 (2003) doi: 10.1007/s10162-002-3055-1.
  27. [27]S. Verhulst, T. Dau, and C. A. Shera, “Nonlinear time-domain cochlear model for transient stimulation and human otoacoustic emission,” The Journal of the Acoustical Society of America 132(6), 3842–3848 (2012) doi: 10.1121/1.4763989.DOI
  28. [28]J. B. Allen and M. M. Sondhi, “Cochlear macromechanics: Time domain solutions,” The Journal of the Acoustical Society of America 66(1), 123–132 (1979) doi: 10.1121/1.383064.DOI
  29. [29]V. Vencovský, A. Vetešník, and A. W. Gummer, “Nonlinear reflection as a cause of the short-latency component in stimulus-frequency otoacoustic emissions simulated by the methods of compression and suppression,” J. Acoust. Soc. Am. 147(6), 3992–4008 (2020) doi: 10.1121/10.0001394.DOI
  30. [30]R. Sisto, C. A. Shera, A. Altoè, and A. Moleti, “Constraints imposed by zero-crossing invariance on cochlear models with two mechanical degrees of freedom,” J. Acoust. Soc. Am. 146(3), 1685–1695 (2019).DOI
  31. [31]R. Sisto, D. Belardinelli, and A. Moleti, “Fluid focusing and viscosity allow high gain and stability of the cochlear response,” J. Acoust. Soc. Am. 150(6), 4283–4296 (2021).DOI
  32. [32]Y. Murakami, “Fast time-domain solution of a nonlinear three-dimensional cochlear model using the fast Fourier transform,” J. Acoust. Soc. Am. 150(4), 2589–2599 (2021).DOI
  33. [33]H. Lamb, Hydrodynamics, 6th ed. (Cambridge University Press, Cambridge, 1932).
  34. [34]G. G. Stokes, “On the effect of the internal friction of fluids on the motion of pendulums,” Trans. Cambridge Philos. Soc. 9(Part II), 8–106 (1851).
  35. [35]R. J. Diependaal, H. Duifhuis, H. W. Hoogstraten, and M. A. Viergever, “Numerical methods for solving one-dimensional cochlear models in the time domain,” The Journal of the Acoustical Society of America 82(5), 1655–1666 (1987) doi: 10.1121/1.395157.DOI
  36. [36]Y. Murakami, “Efficiency limit of nonuniform grid setting in two-dimensional cochlear model,” Acoust. Sci. Tech. 40(5), 336–343 (2019) doi: 10.1250/ast.40.336.DOI
  37. [37]Y. Murakami, “Influence of discretization error on instability of cochlear model,” J. Acoust. Soc. Am. 144(6), 3593–3602 (2018) doi: 10.1121/1.5083836.DOI
  38. [38]K. K. Charaziak and J. H. Siegel, “Estimating cochlear frequency selectivity with stimulus-frequency otoacoustic emissions in chinchillas,” J. Assoc. Res. Otolaryngol. 15, 883–896 (2014) doi: 10.1007/s10162-014-0487-3.DOI
  39. [39]C. A. Shera and J. J. Guinan, “Evoked otoacoustic emissions arise by two fundamentally different mechanisms: A taxonomy for mammalian OAEs,” The Journal of the Acoustical Society of America 105(2), 782–798 (1999) doi: 10.1121/1.426948.DOI
  40. [40]L. Robles and M. A. Ruggero, “Mechanics of the mammalian cochlea,” Physiological Reviews 81(3), 1305–1352 (2001) doi: 10.1152/physrev.2001.81.3.1305.DOI
  41. [41]Y. Murakami and T. Fuji, “Difference between frequency and suppression tuning curves in a two-dimensional cochlear model,” JASA Express Lett. 2, 094402 (2022) doi: 10.1121/10.0013998.DOI
  42. [42]P. van Dijk and H. P. Wit, “Amplitude and frequency fluctuations of spontaneous otoacoustic emissions,” The Journal of the Acoustical Society of America 88(4), 1779–1793 (1990) doi: 10.1121/1.400199.DOI
  43. [43]A. L. Nuttall, K. Grosh, J. Zheng, E. de Boer, Y. Zou, and T. Ren, “Spontaneous basilar membrane oscillation and otoacoustic emission at 15 kHz in a guinea pig,” J. Assoc. Res. Otolaryngol. 5, 337–348 (2004) doi: 10.1007/s10162-004-4045-2.DOI
  44. [44]Y. P. Emelianova, V. V. Emelyanov, and N. M. Ryskin, “Synchronization of two coupled multimode oscillators with time-delayed feedback,” Commun. Nonlinear Sci. Numer. Simul. 19, 3778–3791 (2014) doi: 10.1016/j.cnsns.2014.03.031.DOI
  45. [45]T. J. Kippenberg, R. Holzwarth, and S. A. Diddams, “Microresonator-based optical frequency combs,” Science 332, 555–559 (2011) doi: 10.1126/science.1193968.DOI
  46. [46]A. Hugi, G. Villares, S. Blaser, H. C. Liu, and J. Faist, “Mid-infrared frequency comb based on a quantum cascade laser,” Nature 492(7428), 229–233 (2012) doi: 10.1038/nature11620.DOI
  47. [47]J. B. Khurgin, Y. Dikmelik, A. Hugi, and J. Faist, “Coherent frequency combs produced by self frequency modulation in quantum cascade lasers,” Applied Physics Letters 104(8), 081118 (2014) doi: 10.1063/1.4866868.DOI
  48. [48]E. de Boer and A. L. Nuttall, “Spontaneous basilar-membrane oscillation (SBMO) and coherent reflection,” J. Assoc. Res. Otolaryngol. 7(1), 26–37 (2006) doi: 10.1007/s10162-005-0020-9.DOI
  49. [49]E. M. Burns, “Long-term stability of spontaneous otoacoustic emissions,” The Journal of the Acoustical Society of America 125(5), 3166–3176 (2009) doi: 10.1121/1.3097768.DOI
  50. [50]H. P. Wit and A. Bell, “Something in our ears is oscillating, but what? a modeller’s view of efforts to model spontaneous emissions,” Journal of the Association for Research in Otolaryngology 25(4), 313–328 (2024) doi: 10.1007/s10162-024-00940-7.DOI
  51. [51]51M. A. Cheatham, Y. Zhou, R. J. Goodyear, P. Dallos, and G. P. Richardson, “Spontaneous otoacoustic emissions in TectaY1870C/+ mice reflect changes in cochlear amplification and how it is controlled by the tectorial membrane,” eNeuro 5(6), ENEURO.0314–18.2018 (2018) doi: 10.1523/eneuro.0314-18.2018.DOI

Paper details

vlnc:W_x4d9j55d
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Claude (Anthropic)

View statement
Looking for related papers…
Contents