arXiv:q-bio/0409004v2 [q-bio.NC] 2 Sep 2004 ernta ecie h isn fundamental. missing the noisy third perceives a that drive neuron they auditory two (cor- together left/right and the neuron the pathways), in noisy of detection different each to a responding scenario, drives this stochas- tones ghost In input via signals resonance. we complex tic of paper perception pitch a binaural this the simulations, the of In numerical for of mechanism basis plausible processing. the at of on pitch present, level dichotic higher “virtual” partici- is this a can remains detecting that resonance in question pate stochastic the ghost Thus, whether ear). each in noca,Cie30(08,Beo ie,Argentina. Aires, Buenos (1098), † 300 Chile Antorchas, a.O h te ad ti nw htcom- ( that con- binaurally known two presented its is are when it perceived tones hand, also stituent sig- are other sounds complex the plex the On sim- of a per- nal. presentation of the case monoaural the of ple considering proposal level by original process, peripheral ception The the in system. concentrated the lack nevertheless by that is perceived signals which frequency, harmonic fundamental more the or combi- two a an of the to yields nation response of noise neural of subharmonic basis this level optimal the Under appropriate on an resonance. paradigm, perception proposed stochastic the ghost been for so-called has mechanism pitch a of controver- Recently, remains sys- which auditory of sial. of the perception mechanism by the sounds the tem, complex is of example pitch evolutionary classical the an A have efficiently could payoff. solving problem ner- species, sensory the many for this In task nontrivial system. combination a vous is the inputs from multiple of environmental resulting of signals processing complex and perception The ∗ lcrncades [email protected] address: Electronic lcrncades al.aezeauce;as tFun at also [email protected]; address: Electronic eatmn eF de Departament erlmcaimfrbnua ic ecpinvagots ghost via perception pitch binaural for mechanism neural A os eesehnetersos ftetidnuo tfreq at neuron third the of response the enhance levels noise orsLcrnuo oe ihceia yassexplicit neu synapses third chemical a with to synaptically model applied neuron are Morris-Lecar noise) (plus outputs rvosydsrbdo hs eoac.Frtecs fain a of case the For resonance. ghost of described previously ieetproi inlec wt frequencies (with each signal periodic different a ope onsvagotsohsi eoac.I hssche this In resonance. stochastic ghost via sounds complex o h hr erna frequency a at neuron third the for kf tcatcmcaimcntu rs ihra h periphera the pitch. T at of constant. either perception time the arise synaptic in thus the processing of can function mechanism a stochastic as observed be can epeetapyilgclypasbebnua mechanism binaural plausible physiologically a present We 0 ∆ + ´ sc niyraNcer nvria oi`ciad Ca Polit`ecnica de Universitat Nuclear, Enginyeria e isica f and f 2 ( = k 1) + al Balenzuela Pablo f 0 ∆ + f os sas ent nac h ae fms rbbespik probable most of rates the enhance to seen also is noise ) Dtd coe 8 2018) 28, October (Dated: f r i.e. = daci´on f 0 one ∗ + n od Garc Jordi and k f +1 ∆ 1 f / = 2 inl o htrao,tepthi sal eerdto referred usually is pitch the the in reason, present that spectrally For not is fundamen- signal. frequency the that if is even pitch funda- tal, perceived a the of whose frequency), integers (signals mental multiple par- sounds are the complex frequencies In harmonic constituent tone. of pure quantified case a be with ticular sub- it nevertheless a comparing complex can by is a accurately tones) which pure of attribute, of pitch combination jective perceived a by the (formed contrast, sound In the of [8]. quency. sounds problem complex the of upon pitch light the to new of is neuron shed perception fundamental a has the of which [7] in response missing harmonics the under- of of poorly combination study is a pro- a signals and Recently, combined detection of stood. the type in but this potentials of [6], evoked cessing cortex of visual analysis human the the diagnostic in as for such used hand, purposes, frequency commonly other lower the are signal On a signals frequency to [5]. two-frequency high resonance neuron vibrational a a via of driving that response instance, the for enhances known, is It stochastic and usually conditions. 2] signals, 4] [1, input [3, deterministic simple under to both has neurons harmonic, effort of numerically and response Substantial theoretically the examine manner. to neu- dedicated all-or-none other been from an or in environment rons) the external to from responding (either detectors, signals threshold noisy as ered tbt h eihrladahge processing higher a and peripheral ghost level. the of occurring both mechanism perception at pitch basic explain same can the resonance way, this In nadto,w hwta iia resonances similar that show we addition, In . kf h ecie ic fapr oei ipyisfre- its simply is tone pure a of pitch perceived The signals. input multiple of case the is studied less Much consid- be can neurons sensory conditions many Under 0 and e w ern r rvnb os and noise by driven are neurons two me, .Pthpreto ysnl neurons single by perception Pitch A. ycniee,so htintermediate that show considered, ly o.Ornmrclrsls sn the using results, numerical Our ron. amnccmiaino nus( inputs of combination harmonic f ee ra ihrlvlo neural of level higher a at or level l 2 ece ls to close uencies ´ esgetdghost-resonance-based suggested he ia-Ojalvo ( = o h ecpino h ic of pitch the of perception the for auy,Clm1,E022Tras,Spain Terrassa, E-08222 11, Colom talunya, k +1) .INTRODUCTION I. f † 0 where , f ohsi resonance tochastic 0 si h cases the in as , > k ) n their and 1), f 1 ing = 2 in this case as a “virtual pitch”, and its perception is sentation), and act upon a third neuron which is expected sometimes called the “missing fundamental illusion”. to perceive the pitch of the combined signal. Our results The neural mechanism underlying pitch perception re- show that this higher-level neuron is indeed able to per- mains controversial. From a neurophysiological perspec- ceive the pitch, hence providing a neural mechanism for tive, the perceived pitch is associated with the inter- the binaural experiments. spike interval statistics of the neuronal firings [9, 10]. The analysis presented in Refs. [7, 8] shows that a neu- ron responds optimally to the missing fundamental of II. MODEL DESCRIPTION a harmonic complex signal for an intermediate level of noise, making use of two ingredients: (i) a linear inter- A. Neuron Model ference of the individual tones, which naturally leads to signal peaks at the fundamental frequency, and (ii) a We describe the dynamical behavior of the neurons nonlinear threshold that detects those peaks (with the with the Morris-Lecar model [17], help of a suitable amount of noise, provided the signal is deterministically subthreshold). The behavior of this dVi 1 app ion syn relatively simple model yields remarkably good agree- = (Ii − Ii − Ii )+ Diξ(t) (1) ment with previous psychophysical experiments [11]. The dt Cm ghost stochastic resonance dWi phenomenon has been termed = φΛ(Vi)[W∞(Vi) − Wi] (2) (GSR), and has been replicated experimentally in ex- dt citable electronic circuits [12] and lasers [13]. where Vi and Wi stand for the membrane potential and the fraction of open potasium channels, respectively, and the subindex i labels the different neurons, with i =1, 2 B. Signal integration and processing of distributed representing the two input neurons and i = 3 denoting inputs the processing neuron. Cm is the membrane capacitance app syn per unit area, Ii is the external applied current, Ii Besides the question of how pitch is perceived, an- is the synaptic current, and the ionic current is given by other contested debate relates to where perception takes ion 0 place. Although interval statistics of the neuronal fir- Ii = gCaM∞(Vi)(Vi − VCa)+ ings [9, 10] show that pitch information exists in pe- 0 0 gK Wi(Vi − VK )+ gL(Vi − VL ) (3) ripheral neurons, other results seem to indicate that, at least to some extent, pitch perception takes place at a 0 where ga (a = Ca,K,L) are the conductances and Va higher level of neuronal processing [14]. A typical exam- the resting potentials of the calcium, potassium and leak- ple is found in binaural experiments, in which two com- ing channels, respectively. The following functions of the ponents of a harmonic complex signal enter through dif- membrane potential are also defined: ferent ears. It is known that in that case a (rather weak) low-frequency pitch is perceived. This is called “dichotic 1 V − VM1 pitch”, and can also arise from the binaural interaction M∞(V )= 1 + tanh (4) 2   VM2  between broad-band noises. For example, Cramer and 1 V − VW 1 Huggins [15] studied the effect of a dichotic white noise W∞(V )= 1 + tanh (5) when applying a progressive phase shift across a narrow- 2   VW 2  band of frequencies, centered on 600 Hz, to only one of V − VW Λ(V ) = cosh 1 , (6) the channels. With monaural presentation listeners only  2VW 2  perceived noise, whereas when using binaural presenta- tion over headphones, listeners perceived a 600-Hz tone where VM1, VM2, VW 1 and VW 2 are constants to be spec- against a background noise. ified later. The last term in Eq. (1) is a white noise term It is worth examining whether the ghost resonance of zero mean and amplitude Di, uncorrelated between mechanism introduced by Chialvo et al. [7, 8] can also different neurons. account for the binaural effects described above. Ghost In the deterministic and single-neuron case, this sys- resonance has already been seen to be enhanced by cou- tem shows a bifurcation to a limit cycle for increasing ap- pling in experiments with diffusively coupled excitable plied current Iapp [18]. This bifurcation can be a saddle- lasers [16], but no studies in synaptically coupled neurons node (type I) or a subcritical Hopf bifurcation (type II) have been made so far. Given that chemical synapses depending on the parameters. We chose this last option lead to pulse coupling, a reliable coincidence detection for the numerical calculations presented in this paper. is required in order for ghost resonance to arise in this The specific values of the parameters used in what fol- case. In what follows we examine the situation in which lows are shown in table (I) [19]. The equations were in- two different neurons receive one single component of the tegrated using the Heun method [21], which is equivalent complex signal each (so that each neuron represents de- to a second order Runge-Kutta algorithm for stochastic tection at a different auditory channel in a binaural pre- equations. 3

B. Synapsis model Parameters Morris-Lecar: TII 2 Cm 5 µF/cm 2 In this work we couple the neurons using a simple gK 8 µS/cm 2 model of chemical synapsis [20]. In this model, the synap- gL 2 µS/cm tic current through neuron i is given by 2 gCa 4.4 µS/cm

syn syn VK −80mV Ii = gi rj (Vi − Es), (7) VL −60mV j∈neigh(X i) VCa 120 mV where the sum runs over the neighbors that feed neuron i, VM1 −1.2mV syn gi is the conductance of the synaptic channel, rj stands VM2 18mV for the fraction of bound receptors of the postsynaptic VW 1 2mV channel, Vi is the postsynaptic membrane potential, and VW 2 30mV Es is a parameter whose value determines the type of φ 1/25 s−1 synapsis (if larger than the rest potential, e.g. Es = 0 mV, the synapsis is excitatory; if smaller, e.g. Es = Parameters Synapsis −80 mV, it is inhibitory). α 0.5 ms−1mM−1 The fraction of bound receptors, ri, follows the equa- β 0.1 ms−1 tion gsyn (specified in each case)

dri τsyn (specified in each case) = α[T ]i(1 − ri) − βri , (8) dt Es 0mV i i where [T ]i = θ(T0 +τsyn −t)θ(t−T0) is the concentration TABLE I: Parameters values of the Morris- of neurotransmitter released in the synaptic cleft, α and β Lecar and synapse models used in this work. i are rise and decay time constants, respectively, and T0 is the time at which the presynaptic neuron (labeled now i) fires, what happens whenever the presynaptic membrane potential exceeds a predetermined value, in our case cho- sen to be 10 mV. The time during which the synaptic situation, the input neurons fire exactly with the frequen- connection is active is given by τsyn. The values of the cies at which they are modulated. If f1 = kf0 (or, equiv- parameters that we use, specified below, were taken from alently, T1 = T0/k, where Ti is the period corresponding [20] and could be read from table (I). to the frequency fi) and f2 = (k+1)f0 (T2 = T0/(k+1)), the two input neurons exhibit simultaneous spikes at in- tervals T0 =1/f0 (provided the two harmonic signals are III. THE CASE OF DISTRIBUTED HARMONIC in phase), so that the synaptic current acting on the third COMPLEX SIGNALS neuron has maxima with the same frequency, as can be observed in Fig. 1(c). In this example T1 = 100 ms and As mentioned above, we consider a network of three T2 = 150 ms, so that k = 2 and T0 = 300 ms. Un- neurons organized in two layers. The first layer is com- der these conditions, and for an adequate value of the posed of two units (called “input neurons”) that receive coupling strength gsyn, the processing neuron fires with the external inputs, and whose responses act upon the frequency f0 =1/300 kHz, as shown in Fig. 1(d). Hence, processing layer, composed in this case of only one unit in this case a deterministic ghost resonance is observed. (called “processing neuron”) . The coupling is unidirec- The previous example, however, is not realistic, since tional from each of the input neurons to the processing in normal conditions a neuron is affected by a substan- neuron. Of course, physiological realism dictates that tial level of noise coming from, among other sources, the more than three neurons will be present. However, we background activity of other neurons acting upon it. This model here for simplicity the simplest possible case; one causes a drift in the spike times and a broadening in the can expect that adding more neurons will only improve distribution of inter-spike intervals. We will now show the results. that even in this case the missing fundamental frequency In order to analyze the global response of this network can be successfully detected, as was suggested in [7, 8] for to a distributed complex signal, we apply to each one of a single neuron, even if the synaptic coupling gsyn and the app the input neurons a periodic external current with fre- applied current in the output neuron, I3 , are slightly quencies f1 and f2. In response these neurons emit a below the bifurcation threshold, so that the neuron does sequence of spikes with inter-spike interval distributions not fire in absence of noise (D3 = 0). centered at f1 and f2 and with variances directly related With this in mind, we conduct a series of numerical ex- to the noise intensities D1 and D2. periments looking for the occurrence of ghost stochastic Figure 1 shows the behavior of the system in the ab- resonance. We choose f1 = 2 Hz and f2 = 3 Hz, so the sence of noise (D1 = D2 = D3 = 0). In this deterministic ghost resonance should be located at f0 = 1 Hz. As is 4

150ms 100ms 2000 800 (a) (d) 1500 600 60 (a)60 (b) 40 40 1000 400 > (ms) p

-60 -60 R 0 500 1000 1500 0 500 1000 1500 0.5 pdf time (ms) time (ms) 500 0.4 D3=4 300ms 60 (d) 60 (c) 300ms 0.3 3000 40 0.8 (c) (f) 40 20 3 0.6 200 3 T0 syn V

I 0 0.4 20 T1 fraction 100 -20 T 0.2 2 D3=8 0 -40 0 0 -60 0 2 4 6 8 0 1000 2000 3000 4000 0 500 1000 1500 0 500 1000 1500 time (ms) time (ms) D (noise intensity) Tp (ms)

FIG. 1: Deterministic response to a distributed harmonic FIG. 2: Left panels: response of the processing neuron for complex signal. The membrane potential for the three neu- increasing noise intensity: (a) mean time between spikes hTpi, rons is shown: (a,b) input neurons, (d) processing neuron. (b) normalized variance of the distribution Rp = σp/hTpi, and syn The synaptic current acting on neuron 3, I3 , is shown in (c) fraction of pulses spaced around T0 = 1/f0, T1 = 1/f1 plot (c). The two inputs neurons are fed with two sinusoidal and T2 = 1/f2 as a function of the noise amplitude in the signals of periods T1 = 150 ms and T2 = 100 ms respec- processing neuron, D3. Right panels: probability distribution tively (which gives a ghost resonance of T0 = 300 ms). The functions of the time between spikes Tp for three values of the synaptic coupling between input and processing neurons is noise intensity D3: (d) D3 = 1, (e) D3 = 4, and (f) D3 = 8. gsyn = 2 nS and τsyn = 1 ms. All noise intensities are zero Parameters are τsyn = 35 ms and gsyn = 1 nS for the synapsis D1 = D2 = D3 = 0. and we used f1 = 2 Hz and f2 = 3 Hz (which gives f0 = 1 Hz) for the sinusoidal signals in the input neurons. usual in neurophysiology, in order to quantify the behav- ior of the system, we follow the time between consecutive f1 =1/T1 and f2 =1/T2, as can be observed in Fig. 2(f). spikes, Tp. In what follows, we analyze the first two mo- In the binaural mechanism of ghost stochastic reso- ments of the distribution of Tp, its mean value hTpi and nance described above, synaptic coupling obviously plays its normalized variance Rp = σp/hTpi. To estimate the an important role, since the transfer of the input modu- coherence of the output with the frequencies of interest, lation from the sensory neurons to the processing neuron we also compute the fraction ft0 of inter-spike intervals occurs synaptically. Taking into account that synaptic in the neighborhood of T0 = 1/f0. The dependence of transmission is an intrinsically dynamical phenomenon these variables (corresponding to the processing neuron) (whose temporal behavior we are modelling explicitly), on the noise intensity D3 is shown in Fig. 2. These it is natural to expect that the characteristic time scale results display a clear resonance at D3 ∼ 4. The nor- of this process will influence the occurrence of the reso- malized variance of the inter-spike interval distribution nance. Indeed, the results shown above correspond to an [Fig. 2(b)] exhibits a minimum when the spikes of the optimal value of the synaptic time τsyn. As shown in Fig. third neuron are spaced, on average, hTpi = 1000 ms [Fig. 3 for fixed noise level D3, a resonance in the response of 2(a)]. Additionally, around 80% of the spikes are spaced the system to the missing fundamental is also observed ±5% around T0 = 1000 ms for D3 ∼ 4 [Fig. 2(c)]. These with respect to τsyn. results clearly indicate that noise enhances the response We recall that τsyn represents the time during which of the processing neuron at the frequency f0, which is the neurotransmitters remain in the synaptic cleft before not present in the input neurons. they start to disappear with rate β, and it is a measure The right panels of Fig. 2 show the probability distri- of the width of the pulses of the synaptic current re- bution functions of the inter-spike intervals Tp for three ceived by the processing neuron. Therefore, for low τsyn values of the noise in the processing neuron. For low [Fig. 3(d)] the synaptic pulses are very narrow, and hence noise intensities, the neuron spikes most likely when two coincidence detection is compromised. The characteris- input spikes arrive together, but with randomly one or tic probability distribution function in this case presents more of these coincidence events is skipped. For this rea- peaks at multiples of T0, indicating that even if the noise son, the probability distribution function shows peaks level is optimized, coincident spikes from input neurons centered at multiples of T0, as it usually happens in con- are skipped. ventional stochastic resonance [22]. As the noise level As τsyn increases the current pulses widen and coinci- increases skips occurs less frequently, until an optimal dence detection improves, so that an optimal situation is noise for which almost all spikes occur every T0, i.e. at reached for which the ghost resonance is very clear. But the missing fundamental frequency. For even larger noise if we continue increasing the value of τsyn the synaptic intensities, spikes appear at the original input frequencies pulses become exceedingly wide and sequences of double 5

2000 800 (a) be 1500 600 (d) 1000 400 > (ms) τ f1 = kf0 + ∆f, f2 = (k + 1)f0 + ∆f , (9) p syn=5ms

R 0.8 pdf 500 τ f2 − f1, it should always display a fixed resonance at syn=35ms 0.4 f0, independently of the frequency shift ∆f. But if the 3000 pitch detection does depends on ∆f, it will no longer be 1 (c) (f) 0.8 T0 200 perceived as the difference between the input frequencies. 0.6 T ds This last situation is in fact what was found in human fraction 0.4 100 τ =150ms 0.2 syn experiments [11]. The neural mechanism proposed in [7, 0 0 0 50 100 150 0 1000 2000 3000 4000 8] shows that the frequency of the ghost resonance shifts τ (ms) T (ms) syn p linearly with ∆f following the relation, ∆f FIG. 3: Left panels: (a) mean time between spikes, (b) nor- fr = f0 + . (10) malized variance of the inter-spike interval distribution, and k +1/2 (c) fraction of pulses around T0 = 1/f0 and Tds = 1/fds as a in agreement with the auditory experimental results of function of τsyn. Right panels: Probability distribution func- Refs. [9, 10, 11]. tions of the inter-spike intervals T for three values of τsyn: p We now examine whether a scaling similar to that of (d) τsyn = 5 ms, (e) τsyn = 35 ms, and (f) τsyn = 150 ms. The value of gsyn is different for each value of τsyn ranging Eq. (10) is observed in the response of the processing from gsyn = 2.50 nS for τsyn = 1.5 ms up to gsyn = 1.00 nS neuron. We fix the noise intensity D3 and synaptic time for τsyn = 150 ms. f1 = 2 Hz and f2 = 3 Hz (which gives τsyn to their optimal values at the resonance (D3 = 4, f0 = 1 Hz) for the sinusoidal signals in the input neurons was τsyn = 35ms) and compute the probability of observing a used. spike with rate fr for increasing ∆f. The results are plot- ted (in gray scale) in Fig. 5 as a function of f1, and show that the largest probability corresponds to spike rates fol- spikes appear (spaced by Tds = 1/fds). This happens lowing the prediction of relation relation (10). Changing because noise can excite two spikes while the synaptic the noise intensity only obscures the observation of the current remains high. Indeed, Fig. 3(c) shows that the spike density, but it does not affect the agreement with fraction of spikes occurring at intervals around Tds (±5%) the theoretical expression. In the bottom of Fig. 5 one begins to be important for τsyn > 50 ms. The corre- can also see traces of less probable spikes, corresponding sponding distribution function in Fig. 3(f), shown here to a trivial subharmonic response of the system. for τsyn = 150 ms, corroborates this fact. Figure 5 shows that the processing neuron emits spikes The joint effect of the synaptic time τsyn and the noise following Eq. (10) for k =2, 3, 4, 5. As mentioned above, level D3 can be observed in the three-dimensional plots this relation is sustained by experimental data of pitch shown in Fig. 4. This figure shows Rp and ft0 as a detection [11]. Those experimental results indicate that function of both D3 and τsyn. We can see the response equidistant tones in monoaural presentation do not pro- of the processing neuron at the missing fundamental is duce constant pitch, similarly to what we observe in our most favorable when both parameters are simultaneously binaural numerical experiments. We are not aware of optimized. The normalized variance of the inter-spike binaural human experiments shifting the frequency com- interval distribution, Rp, shows a clear minimum for ponents as in [11], which would be interesting to compare τsyn ∼ 35 ms and D ∼ 4. For these same parameter with our numerical predictions in Eq. (10). values, the fraction of spikes ft0 occurring at intervals around T0 exhibits a maximum at almost 80%. V. CONCLUSIONS

In this paper we demonstrate the phenomenon of ghost IV. THE INHARMONIC CASE stochastic resonance in a neural circuit where two neu- rons receive two components of a complex signal and A paradigmatic experimental result in pitch perception their outputs drive a third neuron that processes the in- refers to the pitch reported by human subjects to the formation. The results show that the processing neu- presentation of a inharmonic complex sound, in which ron responds preferentially at the “missing fundamen- the originally harmonic components of the input are all tal” frequency, and that this response is optimized by shifted in frequency by a constant ∆f. In such a way the synaptic noise and by synaptic time constant. The pro- individual component are still separated in frequency by cessing neuron is able to detect the coincident arrival of a constant missing “fundamental” f0, but are no longer spikes from each of the input neurons, and this coinci- multiples of it. The frequencies f1 and f2 are chosen to dence detection is analogous to the linear interference of 6

1 0.8 0.6 t0 f 0.4

0.5 p R 0.2 0 0 9 9 8 7 150 8 7 150 6 120 6 120 5 90 5 90 4 60 4 60 3 2 30 3 2 30 D 1 0 τ (ms) D 1 0 τ (ms) 3 syn 3 syn

FIG. 4: Left: fraction of pulses ft0 occurring at intervals Tp equal (±5%) to the period of the ghost resonance (T0 = 1/f0). Right: normalized variance of the inter-spike interval distribution, Rp. Both quantities plotted as function of noise intensity (D3) and τsyn.

2 in the single-neuron case [7]. A brain structure candidate 1.8 for this dynamics is the inferior colliculus, which receives 1.6 multiple inputs from a host of more peripheral auditory nuclei. Details of the physiology of this nucleus are still 1.4 uncertain, but enough evidence suggests that temporal 1.2 and frequency representations of the inputs are present

(Hz) 1

r in the spike timing of their neurons. Our results suggest

f 0.8 that the neurons in this nucleus can exhibit the dynam- 0.6 ics described here, thus participating in the perception 0.4 of binaural pitch. The main consequence of these obser- 0.2 vations is that pitch information can be extracted mono 0 or binaurally via the same basic principle, i.e. ghost 2 2.5 3 3.5 4 4.5 5 stochastic resonance, operating either at the periphery f (Hz) 1 or at higher sensory levels. Acknowledgments FIG. 5: Probability of observing a spike in the processing neuron with instantaneous rate fr (in gray scale) as a function of the frequency f1 of one of the input neurons. We can observe a remarkable agreement of the responses following the We thank Dante R. Chialvo for guidance and useful lines predicted by Eq. (10) for k = 2, 3, 4, 5 (dashed lines from comments on the manuscript. We acknowledge financial top to bottom). Parameters: τsyn = 35 ms, gsyn = 1.0 nS, support from MCyT-FEDER (Spain, projects BFM2002- D3 = 4.0, f1 = 2Hz +∆f, f2 = 3Hz +∆f. 04369 and BFM2003-07850), and by the Generalitat de Catalunya. P.B. acknowledges financial support from the Fundaci´on Antorchas, Chile 300 (1098), Buenos Aires, harmonic components responsible of the ghost response Argentina.

[1] P. Parmananda, C. H. Mena, and G. Baier, Phys. Rev. G. V. Savino, Phys. Rev. E 65, 050902(R) (2002). E 66, 047202 (2002). [8] D. R. Chialvo, Chaos 13, 1226 (2003). [2] C.R. Laing and A. Longtin, Phys. Rev. E 67, 051928 [9] P. A. Cariani and B. Delgutte, J. Neurophysiol. 76, 1698– (2003). 1716 (1996). [3] A. Longtin and D.R. Chialvo, Phys. Rev. Lett. 81, 4012 [10] P. A. Cariani and B. Delgutte, J. Neurophysiol. 76, 1717– (1998). 1734 (1996). [4] B. Lindner, J. Garcia-Ojalvo, A. Neiman, and L. [11] J.F. Schouten, R.J. Ritsma, and B.L. Cardozo, J. Acous- Schimansky-Geier, Phys. Rep. 392, 321 (2004). tic Soc. Am. 34, 1418–1424 (1962). [5] E. Ullner, A. Zaikin, J. Garc´ia-Ojalvo, R. B´ascones, and [12] O. Calvo and D.R. Chialvo, “Ghost stochastic resonance J. Kurths, Phys. Lett. A 312, 348 (2003). on an electronic circuit”, unpublished (2004). [6] J. D. Victor and M. M. Conte, Visual Neurosci. 17, 959 [13] J.M. Buld´u, D.R. Chialvo, C.R. Mirasso, M.C. Torrent, (2000). and J. Garc´ıa-Ojalvo, Europhys. Lett. 64, 178 (2003). [7] D. R. Chialvo, O. Calvo, D. L. Gonzalez, O. Piro, and [14] C. Pantev, T. Elbert, B. Ross, C. Eulitz, and E. Terhardt, 7

Hearing Research 100, 164–170 (1996). [19] K. Tsumoto, T. Yoshinaga, K. Ahiara, and H. Kawakami, [15] E. C. Cramer and W. H. Huggins, J. Acoust. Soc. Am. in Proc. International Symposium on Nonlinear Theory 30, 858–866 (1958). and its Applications 1, 363 (2002). [16] J.M. Buld´u, C. M. Gonz´alez, J. Trull, M.C. Torrent, and [20] A. Destexhe, Z. F. Mainen, and T. J. Sejnowski, Neural J. Garc´ıa-Ojalvo, “Coupling-mediated ghost resonance in Comp. 6, 14 (1994). mutually injected lasers,” unpublished (2004). [21] J. Garc´ıa-Ojalvo and J. M. Sancho, Noise in Spatially [17] C. Morris and H. Lecar, Biophys. J. 35, 193 (1981). Extended Systems (Springer, New York, 1999). [18] M. St-Hilaire A. Longtin, J. Comp. Neurosc., 16, 299- [22] A. Longtin, J. Stat. Phys. 70, 309 (1993). 313, (2004).