A question about RMS value of quantization error

when the input signal obey bennett rule ( Probability density function for the quantization error is constant) ,the RMS of quantization error is Vlsb/square root of 12.
When input signal is a sinewave ( Probability density function for the quantization error not constant), I think the RMS of quantization error is not Vlsb/square root of 12 , maybe Vlsb/square root of 8.
Is that right?

RMS 值取决于输入信号的幅值概率密度分布，如果该正弦波破坏了幅值的等概率分布就不是你上面说的那个值

The RMS approaches the value of # Vlsb/sqrt(12) # when the resolution N becmoes higher. It is a approximation anyway, because it is conditioned that "Probability density function for the quantization error is constant", which is not exact as that in reality.

那么，是否可以认为极端情况（ADC只有一位,且基本上输入正弦波幅度满足其满量程时），对于正弦波的量化误差RMS值为 Vlsb/sqrt(8)。
如果上面成立的话，对于一些噪声整形滤波器（西格玛-delta）来说（实际比较位数很少），是否传统的ADC SNR 计算就不对了？
谢谢。