5D Library › 节律声刺激、听觉稳态反应与双耳等时调制工程 · English
第二章 听觉稳态反应与阿诺德舌
「夹带」的两种意思
说一个节律声「夹带」了脑波,可以指两件很不同的事。弱的意思是诱发反应:每个周期的声音都诱发一个小的神经反应,这些反应叠加起来,在脑电的刺激频率处形成一条谱线,即听觉稳态反应(ASSR)。40 Hz 附近的稳态反应尤其明显 (Galambos et al. 1981),在听力学中已是成熟的客观测听手段 (Picton et al. 2003)。强的意思是同步:一个本来就存在的内在节律被外部驱动拉到驱动的频率上,并在驱动停止后仍保持一段时间。后者需要一个振荡器,前者不需要。
阿诺德舌
同步的最简单模型是 Adler 方程 (Adler 1946):设内在节律与驱动之间的相位差为 ,失谐为 ,耦合强度为 ,则 。没有噪声时,当且仅当 时相位差停在一个固定值上,这就是锁定;在失谐与耦合构成的平面上,锁定区是一个从零失谐处张开的「舌」(Kuramoto 1984; Pikovsky et al. 2001)。加入噪声后,锁定变成程度问题,用锁相值(PLV)度量 (Lachaux et al. 1999)。
推导
# Chapter 2. p03: The Arnold tongue of a noisy phase oscillator.
# Adler's equation for the phase difference phi between an intrinsic
# rhythm and a periodic drive, with detuning nu (Hz) and coupling K
# (rad/s), plus white phase noise of strength D:
# dphi = (2 pi nu - K sin phi) dt + sqrt(2 D dt) N(0, 1).
# Without noise the rhythm locks exactly when |2 pi nu| <= K (the
# tongue). With noise, locking is graded; we measure it by the phase
# locking value PLV = |mean exp(i phi)| over 60 s after a 5 s burn-in.
# Euler-Maruyama step 1 ms, fixed seed. D = 2 rad^2/s throughout.
import cmath
import math
import random
def plv(nu, K, D=2.0, T=60.0, dt=0.001, seed=7):
rng = random.Random(seed)
phi, acc, n = 0.0, 0j, 0
burn = int(5.0 / dt)
sd = math.sqrt(2 * D * dt)
w = 2 * math.pi * nu
for i in range(burn + int(T / dt)):
phi += (w - K * math.sin(phi)) * dt + sd * rng.gauss(0, 1)
if i >= burn and i % 10 == 0:
acc += cmath.exp(1j * phi)
n += 1
return abs(acc) / n
if __name__ == "__main__":
nus = (0.0, 0.5, 1.0, 2.0, 4.0)
Ks = (2.0, 6.0, 12.0, 25.0)
print(" PLV after 60 s; rows: coupling K (rad/s); cols: detuning nu")
print(" K \\ nu " + "".join("{:>9.1f}".format(v) for v in nus))
for K in Ks:
row = [plv(v, K) for v in nus]
print(" {:6.1f} ".format(K) + "".join(
"{:9.3f}".format(r) for r in row))
print("\n noiseless tongue edge |nu| = K / (2 pi):")
for K in Ks:
print(" K = {:5.1f} -> |nu| <= {:.3f} Hz".format(
K, K / (2 * math.pi)))
print("\nReading: a weak drive locks a rhythm only if it is almost")
print("at the same frequency already; the tongue widens in")
print("proportion to the coupling. Outside the tongue the PLV")
print("falls towards the chance floor for 6,000 samples.")
# PLV after 60 s; rows: coupling K (rad/s); cols: detuning nu
# K \ nu 0.0 0.5 1.0 2.0 4.0
# 2.0 0.544 0.287 0.146 0.077 0.043
# 6.0 0.838 0.736 0.496 0.251 0.115
# 12.0 0.921 0.917 0.880 0.588 0.244
# 25.0 0.962 0.962 0.960 0.954 0.690
#
# noiseless tongue edge |nu| = K / (2 pi):
# K = 2.0 -> |nu| <= 0.318 Hz
# K = 6.0 -> |nu| <= 0.955 Hz
# K = 12.0 -> |nu| <= 1.910 Hz
# K = 25.0 -> |nu| <= 3.979 Hz
#
# Reading: a weak drive locks a rhythm only if it is almost
# at the same frequency already; the tongue widens in
# proportion to the coupling. Outside the tongue the PLV
# falls towards the chance floor for 6,000 samples.耦合为 6 rad/s 时,零失谐的锁相值为 0.838,失谐 2 Hz 时降到 0.251;无噪声的舌边在 0.955 Hz。耦合 25 rad/s 时舌边在 3.979 Hz,失谐 4 Hz 处锁相值仍有 0.690。舌的宽度与耦合成正比:弱驱动只能锁定频率本来就很接近的节律。这对节律音频的含义是:要把一个 10 Hz 的内在节律拉到 6 Hz,驱动必须非常强;而第一章表明,双耳拍恰恰是弱驱动。
需要记录多长
即使只要求「诱发反应」,也要问它能否被看见。设反应幅度在 40 Hz 附近最大,背景脑电的功率谱密度按 下降。时长为 的傅里叶变换把反应放进宽度为 的一个频点,噪声功率随 减小,于是功率信噪比与 成正比,幅度信噪比与 成正比。程序要求信噪比达到 10 dB,并至少包含十个完整周期。
推导
# Chapter 2. p04: How long must a rhythm play before EEG can see it?
# Illustrative model of an auditory steady-state response (ASSR):
# response amplitude a(fm) = 0.15 + 0.35 exp(-((fm - 40) / 10)^2) uV
# for a fully modulated (isochronic) drive, scaled by the drive;
# background EEG power density P(f) = 4 / f uV^2/Hz (a 1/f law).
# A DFT of T seconds puts the response in one bin of width 1/T, so the
# power SNR is (a^2 / 2) T / P(fm): it grows in proportion to T and the
# amplitude SNR as sqrt(T). We ask for 10 dB and also for at least ten
# full cycles. Binaural drive = 0.13116 (carrier 400 Hz, from p02).
# Then: the drive of a static tone, wavelengths of 432 and 528 Hz,
# and the chance level of the phase locking value over M epochs.
import math
def amp(fm):
return 0.15 + 0.35 * math.exp(-((fm - 40.0) / 10.0) ** 2)
def t_needed(fm, drive, target_db=10.0):
a = amp(fm) * drive
snr_per_s = (a * a / 2.0) / (4.0 / fm)
t_snr = 10 ** (target_db / 10.0) / snr_per_s
return max(t_snr, 10.0 / fm)
def fmt(t):
if t < 60:
return "{:8.1f} s".format(t)
if t < 3600:
return "{:8.1f} m".format(t / 60)
return "{:8.1f} h".format(t / 3600)
if __name__ == "__main__":
print(" rate Hz a (uV) isochronic T binaural T ratio")
for fm in (0.1, 2.0, 6.0, 10.0, 40.0):
ti, tb = t_needed(fm, 1.0), t_needed(fm, 0.13116)
print(" {:7.1f} {:8.3f} {} {} {:6.1f}".format(
fm, amp(fm), fmt(ti), fmt(tb), tb / ti))
print("\n amplitude SNR at 40 Hz, isochronic, as T grows:")
for T in (1, 4, 16, 64):
s = (amp(40) ** 2 / 2) * T / (4.0 / 40)
print(" T = {:3d} s power {:7.2f} amplitude {:6.2f}".format(
T, s, math.sqrt(s)))
print("\n static tones: modulation depth 0, so drive 0 and a")
print(" response of 0 uV at every rate and every duration.")
for f in (432, 528):
print(" {} Hz: wavelength {:.3f} m in air, {:.3f} m in water,"
.format(f, 343.0 / f, 1480.0 / f))
print(" {:.2f} billion times the 2 nm width of DNA"
.format(1480.0 / f / 2e-9 / 1e9))
print("\n chance PLV for M random-phase epochs, sqrt(pi / (4 M)):")
for M in (10, 40, 160, 640):
print(" M = {:4d} {:.4f}".format(M, math.sqrt(math.pi / 4 / M)))
# rate Hz a (uV) isochronic T binaural T ratio
# 0.1 0.150 9.9 h 574.1 h 58.1
# 2.0 0.150 29.6 m 28.7 h 58.1
# 6.0 0.150 9.9 m 9.6 h 58.1
# 10.0 0.150 5.9 m 5.7 h 58.1
# 40.0 0.500 8.0 s 7.8 m 58.1
#
# amplitude SNR at 40 Hz, isochronic, as T grows:
# T = 1 s power 1.25 amplitude 1.12
# T = 4 s power 5.00 amplitude 2.24
# T = 16 s power 20.00 amplitude 4.47
# T = 64 s power 80.00 amplitude 8.94
#
# static tones: modulation depth 0, so drive 0 and a
# response of 0 uV at every rate and every duration.
# 432 Hz: wavelength 0.794 m in air, 3.426 m in water,
# 1.71 billion times the 2 nm width of DNA
# 528 Hz: wavelength 0.650 m in air, 2.803 m in water,
# 1.40 billion times the 2 nm width of DNA
#
# chance PLV for M random-phase epochs, sqrt(pi / (4 M)):
# M = 10 0.2802
# M = 40 0.1401
# M = 160 0.0701
# M = 640 0.0350满深度的 40 Hz 等时音只需 8.0 s;同频率的双耳拍需要 7.8 min,长 58.1 倍。在 10 Hz,等时音需要 5.9 min,双耳拍需要 5.7 h;在 0.1 Hz,即使等时音也要 9.9 h,此时「十个周期」本身就要 100 s,而低频的脑电背景噪声极大。幅度信噪比随时长按平方根增长:1 s 时为 1.12,16 s 时为 4.47。
静态的 432 Hz 或 528 Hz 纯音调制深度为 0,驱动为 0,任何时长都不会产生节律反应。它们在空气中的波长分别为 0.794 m 与 0.650 m,在水中为 3.426 m 与 2.803 m,分别是 DNA 双螺旋宽度(约 2 nm)的 倍与 倍。声波在这种尺度上对分子只是一个均匀的压力变化,谈不上「共振修复」。
最后,锁相值本身有机会水平: 段随机相位的单位向量,其平均长度约为 。10 段时为 0.2802,640 段时为 0.0350。报告锁相值而不报告段数,等于不报告任何东西。
假说 · 尚未验证
可检验的推论。 若 432 Hz 与 528 Hz 纯音对脑电有频率特异的作用,则在双盲条件下,与 440 Hz 纯音相比,二者应在预先指定的频段与电极上产生不同的功率或锁相。什么会证明这一推论错:若在足够的样本与预先登记的分析下,三个纯音的脑电指标没有差别,则「特殊频率」的说法没有生理基础。
小结
可以测量的「夹带」是稳态反应,它在 40 Hz 附近最强,需要几秒到几分钟的记录才能被看见;双耳拍需要的时间长得多。真正的同步受阿诺德舌约束:弱驱动只能锁定频率接近的节律。静态纯音没有节律可供诱发或同步。
参考文献
Fan, Y. (2026). Rhythmic Acoustic Stimulation, Auditory Steady-State Responses and Modulation Signal Engineering: Binaural versus Isochronic Envelope Depth, 40-Hz Phase-Locking Bounds and Pink-Noise Spectral Masking [节律声刺激、听觉稳态反应与双耳等时调制工程] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090211 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).