Solved! Go to Solution.
Resolution is an important specification for an ADC, it describes the minimum ADC recognition for the analog input variation. For example, the ADC is 12bit, we say the resolution of the ADC is 12bit. But for an actual ADC, because of various reasons, it can not recognize the analog input variation as it’s resolution describe, so we should define the ENOB parameter to describe the it’s actual resolution.
2.4.1 how to derivate ENOB parameterFrom mathematical view, quantization process is to map the continuous analog signal to a discrete definite number set, quantization error is introduced in the process, we can use the quantization error to measure the difference between the actual analog value ax(k) corresponding one ADC sample and approximated digital value {dx(k)},the difference is expressed as d{dx(k),ax(k)},the commonly used quantization error is the root-mean-squared error defined as d{dxk,axk}=( dxk,axk)2.
For a general ADC, we assume the analog signal is Uniformly Distributed, so the uniform quantization is the optimum quantization.
For a general ADC, supposing that the analog signal ranges from -V to V, the nominal resolution is L bit, the desired number of level is 2L,the quantization gap is 2*V/2L,Supposing the sample frequency is Fs, and the signal frequency is F, for meeting the Shannon sampling theorem, the Fs should be greater that 2*F.
Supposing we quantize SINE waveform signal with the ADC, the amplitude of the Sine signal is V.
In an ideal ADC converter, the quantization error is uniformly distributed between -1/2LSB and 1/2LSB,
The sum of root-mean-squared Quantization error (or the energy of the quantization error) or the noise is
In the above formula, x denotes the quantization error between the continuous analog signal and the digital sample, p(x) is the possibility of the quantization error assumed to be uniformly distributed, so
and =2*V/2L
so the noise energy is
For a sine signal with amplitude V, the signal energy is ,
So 10*log(Signal energy/Noise energy)=10*log(/J)=6.02L+1.76
So the ENOB L={10*log(Signal energy/Noise energy)-1.76}/6.02.
If we know the signal energy and the noise energy,using above formula we can get the ENOB L.
2.4.2 one method to compute ENOB parameter
From the formula L=(log(Signal/Noise)-1.76)/6.02. If we can measure the signal and noise energy, then we can get L, L is the ENOB in the data sheet.
Suppose we input a arbitrary initial phase sine signal to the ADC, which cab be expressed as A*sin(wt)+B*cos(wt),the amplitude of the sine signal is , assume the frequency of the signal is FREQ, the signal energy can be expressed as , if we sample the sine signal using sample Frequency FS, and let the FS/FREQ=integer for example 100, in order to increase accuracy.
Assume we get one whole cycle sine discrete signal y(K) by sampling the sine signal, the number of point is NP which equal to FS/FREQ, you can use the formula to get Noise.
K=NP-1
J= SUM[ y(k)-A*sin(k/NP)-B*cos(k/NP)]**2
k=0
we can estimate the optimum parameter A and B, then compute the noise J. let the following partial derivatives equal to 0
d(J)/d(A)=0
d(J)/d(B)=0
we can get the equation
SUM[ y(k)-A*sin(k/NP)-B*cos(k/NP)]*sin(k/NP)=0
SUM[ y(k)-A*sin(k/NP)-B*cos(k/NP)]*cos(k/NP)=0
From the two equation, we can get the A and B value.
k=NP-1
Noise energy = SUM[ y(k)-A*sin(k/NP)-B*cos(k/NP)]**2
k=0
and the Signal energy =(A**2+B**2)/2
From the formula L=(10*log(Signal/Noise)-1.76)/6.02,we can calculate the L which is ENOB defined in data sheet.
We can also calculate ENOB by doing FFT of y(k) sequence, they should have the same result.
Resolution is an important specification for an ADC, it describes the minimum ADC recognition for the analog input variation. For example, the ADC is 12bit, we say the resolution of the ADC is 12bit. But for an actual ADC, because of various reasons, it can not recognize the analog input variation as it’s resolution describe, so we should define the ENOB parameter to describe the it’s actual resolution.
2.4.1 how to derivate ENOB parameterFrom mathematical view, quantization process is to map the continuous analog signal to a discrete definite number set, quantization error is introduced in the process, we can use the quantization error to measure the difference between the actual analog value ax(k) corresponding one ADC sample and approximated digital value {dx(k)},the difference is expressed as d{dx(k),ax(k)},the commonly used quantization error is the root-mean-squared error defined as d{dxk,axk}=( dxk,axk)2.
For a general ADC, we assume the analog signal is Uniformly Distributed, so the uniform quantization is the optimum quantization.
For a general ADC, supposing that the analog signal ranges from -V to V, the nominal resolution is L bit, the desired number of level is 2L,the quantization gap is 2*V/2L,Supposing the sample frequency is Fs, and the signal frequency is F, for meeting the Shannon sampling theorem, the Fs should be greater that 2*F.
Supposing we quantize SINE waveform signal with the ADC, the amplitude of the Sine signal is V.
In an ideal ADC converter, the quantization error is uniformly distributed between -1/2LSB and 1/2LSB,
The sum of root-mean-squared Quantization error (or the energy of the quantization error) or the noise is
In the above formula, x denotes the quantization error between the continuous analog signal and the digital sample, p(x) is the possibility of the quantization error assumed to be uniformly distributed, so
and =2*V/2L
so the noise energy is
For a sine signal with amplitude V, the signal energy is ,
So 10*log(Signal energy/Noise energy)=10*log(/J)=6.02L+1.76
So the ENOB L={10*log(Signal energy/Noise energy)-1.76}/6.02.
If we know the signal energy and the noise energy,using above formula we can get the ENOB L.
2.4.2 one method to compute ENOB parameter
From the formula L=(log(Signal/Noise)-1.76)/6.02. If we can measure the signal and noise energy, then we can get L, L is the ENOB in the data sheet.
Suppose we input a arbitrary initial phase sine signal to the ADC, which cab be expressed as A*sin(wt)+B*cos(wt),the amplitude of the sine signal is , assume the frequency of the signal is FREQ, the signal energy can be expressed as , if we sample the sine signal using sample Frequency FS, and let the FS/FREQ=integer for example 100, in order to increase accuracy.
Assume we get one whole cycle sine discrete signal y(K) by sampling the sine signal, the number of point is NP which equal to FS/FREQ, you can use the formula to get Noise.
K=NP-1
J= SUM[ y(k)-A*sin(k/NP)-B*cos(k/NP)]**2
k=0
we can estimate the optimum parameter A and B, then compute the noise J. let the following partial derivatives equal to 0
d(J)/d(A)=0
d(J)/d(B)=0
we can get the equation
SUM[ y(k)-A*sin(k/NP)-B*cos(k/NP)]*sin(k/NP)=0
SUM[ y(k)-A*sin(k/NP)-B*cos(k/NP)]*cos(k/NP)=0
From the two equation, we can get the A and B value.
k=NP-1
Noise energy = SUM[ y(k)-A*sin(k/NP)-B*cos(k/NP)]**2
k=0
and the Signal energy =(A**2+B**2)/2
From the formula L=(10*log(Signal/Noise)-1.76)/6.02,we can calculate the L which is ENOB defined in data sheet.
We can also calculate ENOB by doing FFT of y(k) sequence, they should have the same result.