Mixed Signal Circuit Design -ADC Part 2
Oct 6, 2023
In ADC specification, we discussed about quantization error.
Next is differential non-linearity.
This differential non-linearity is the same as what we learned in DAC.
For ADC, differential non-linearity is the difference between actual code width of the non-ideal converter and the ideal case.
In this figure, we have given a staircase and the difference between the code width of the non-ideal converter and the ideal case.
Here, we have given the difference between the code width of the non-ideal converter and the ideal case.
We have drawn this to find differential non-linearity.
Now, we have drawn quantization error as we found earlier.
I have already uploaded the video on how to find quantization error.
To find quantization error, we have to reduce the value of the staircase from the input.
To find quantization error, we have to reduce the staircase value from the input.
When we reduce the staircase value, it increases linearly like this.
Similarly, we have drawn quantization error by adding all other values.
In this figure, we have drawn quantization error as before.
Similarly, we have drawn differential non-linearity by reducing the code width of the non-ideal converter and the ideal case.
Now, let us see how to find differential non-linearity error.
Differential non-linearity error is the difference between actual step width minus ideal step width.
The ideal step width is 1 LSB.
Next, we have to find the differential non-linearity error.
In the case of DNL0, the differential non-linearity error is 0.
In this figure, the non-linearity error is 0.5.
In the case of DNL0, the differential non-linearity error is 0.5.
Now, let us see the other cases.
In the case of DNL2, the differential non-linearity error is 1.5 LSB minus ideal step width.
In the case of DNL3, the differential non-linearity error is 0.5 LSB minus ideal step width.
In the case of DNL5, the differential non-linearity error is 0.5 LSB minus ideal step width.
In the case of DNL6, the differential non-linearity error is 0.5 LSB minus ideal step width.
In the case of DNL7, the differential non-linearity error is 0.5 LSB minus ideal step width.
These are the missing codes.
This is a transfer curve for a non-ideal 3-bit DAC with missing code.
Let's see where the missing code comes from.
All the values have a corresponding step width.
But 101 does not have a step width.
Let's consider 101 as a missing code.
In the case of 101, the DNL value is actual step width minus ideal step width.
In the case of DNL5, the DNL value is actual step width minus ideal step width.
In the case of 010, the DNL value is 2 LSB.
In the case of 010, the DNL value is actual step width minus ideal step width.
In the case of 011, the DNL value is 0.5 LSB minus ideal step width.
Note that even if the step width is greater than 1 LSB, there is no guarantee that there is a missing code.
But if the step width is less than 1 LSB, there is a guarantee that there is a missing code.
This is the explanation of the missing code.
This is the integral non-linearity error.
We have studied the integral non-linearity error in the DAC.
The same is the integral non-linearity error in this case.
What did we do to find the integral non-linearity error?
We draw a straight line with two end points and the first and last of the co-transition.
From that, we find how much the transition changes.
This is called the integral non-linearity error.
Now let's move on to the figure and its explanation.
We have given this figure to represent an integral non-linearity.
We have given a non-ideal ADC transfer curve.
This is the quantization error.
We have studied how to draw the quantization error earlier.
This is how this signal is drawn.
We have calculated the quantization error.
Now, let's find the integral non-linearity error.
The integral non-linearity error is defined as how much the transition changes from the ideal ADC.
Now, let's find the difference between 0, 1, 1 and INL3.
INL3 is 3 by 8.
But the transition is slightly different.
There is a difference between these two.
Let's find INL3.
3 by 8 minus 5 by 16.
This value is 5 by 16.
INL3 is 3 by 8 minus 5 by 16.
This is equal to 1 by 16.
We have found the difference.
1 by 8 is 1 LSB.
1 by 16 is 0.5 LSB.
INL3 is 0.5 LSB.
INL error in 4 is 0.
INL error in 5 is 0.
INL error in 6 is equal to minus 0.5.
This is how we find INL error.
The next ADC specification is Offset and Gain Error.
Offset and Gain Error is the same as we discussed in the DAC case.
Let's see how Offset Error occurs.
When there is a difference between the value of 1st code transition and ideal value of 0.5 LSB.
Offset Error is always a constant value.
Condensation Error is always a constant value.
This figure represents Offset Error.
This is the ideal case figure.
This is the ideal case transfer curve.
If there is any offset value, this curve will shift to this side.
This is the same curve.
Condensation Error is always a constant value.
Offset voltage varies from 0.5 to 2.8.
After 2.8, Offset Voltage is a constant value.
The reason for the shift from 0 to 2.8 is the presence of Offset Voltage.
Let's see what Gain Error is.
Gain Error is also known as Scale Factor Error.
Gain Error is the difference in slope of straight line drawn through transfer characteristics and slope 1 of an ideal ADC.
Gain Error is the difference between ideal ADC slope and actual ADC slope.
Let's see the aliasing effect.
Sampling quantization is the process of ADC analog to digital conversion.
Aliasing effect is related to sampling.
There is a criterion for sampling a signal.
It is called Nyquist Criterion.
Nyquist Criterion is that the sampling frequency should always be more than 2 times the message frequency.
Fs should be greater than or equal to 2Fm.
Only if the sampling frequency is more than 2Fm, the signal can be reconstructed.
If the sampling frequency is less than the message frequency, the effect is called aliasing.
Fs should be less than 2Fm.
The different frequency signal is an alias of original signal and its frequency can be calculated using.
To calculate the frequency of an alias signal, Falias is equal to FHL plus Fsample.
FHL is analog signal.
Sample is the sampling frequency.
Falias is the frequency of the alias signal.
Let's see this figure.
This is the figure of aliasing through undersampling.
This is the actual signal.
The sample points are this point, this point, this point, and this point.
If we add these sample points and draw, we will get a signal like this.
If you notice, the frequency of this signal is less than the actual signal.
If the sampling frequency is less than the actual signal, the effect is called aliasing.
To find the aliasing, Falias is equal to FHL plus K into Fsample.
K is equal to minus 2, minus 1, 0, 1, 2, 3, etc.
K is equal to minus infinity to plus infinity.
The peculiarity of this figure is that the frequency of the sample points is taken with a good distance.
But when we reconstruct it, we do not get this signal.
That is the problem of aliasing.
Next, let's see this figure.
This is the process that represents the signal in the time domain and frequency domain.
This is the analog signal.
This is the representation in the frequency domain.
Frequency domain is the frequency of omega.
Time domain is f of t.
These are the techniques used to convert the signal in the frequency domain to the time domain.
Laplace transform, Fourier transform, etc. are techniques used to convert the signal in the time domain to the frequency domain.
This is the frequency domain.
When we do sampling, we get the sampled signal when we multiply the signal in the time domain by a sampling function or a pulse train.
This is the representation in the corresponding frequency domain.
When the sampling frequency is less than the sampling frequency, what happens to the signal?
This is called aliasing.
If the signal in this region is lost, it will be difficult to reconstruct the signal.
This is the process of aliasing.
You have learned the multiplication in the time domain as a property of Fourier transform and Z transform.
We have also learned that the multiplication in the time domain is equivalent to the convolution in the frequency domain.
What can we do to eliminate aliasing?
Aliasing can be eliminated by both sampling at high frequency and by filtering analog signal before sampling and removing any frequencies that are greater than 1.5.
To eliminate aliasing, we sample at high frequency.
After sampling at high frequency, what can we remove?
We filter the analog signal and remove any frequencies that are greater than 1.5.
If we remove any frequencies that are greater than 1.5, we can avoid aliasing.
Similarly, one could also filter the signal post-sampling and eliminate frequencies for which overlap occurs.
We can avoid overlapping by using a filter in post-sampling.
The next property of ADC is signal-to-noise ratio.
Signal-to-noise ratio is the ratio of signal power and noise power.
Signal-to-noise ratio represents the value of RMS input signal into the converter over the RMS value of noise.
That is 20 log VIN divided by V noise is the signal-to-noise ratio.
The equation to find the maximum value of VIN is VIN maximum is equal to V reference divided by 2 root 2.
Instead of V reference, we can write 2 raise to n VLSB.
The equation for quantization error is 1 by VLSB into integral minus 0.5 VLSB to 0.5 VLSB into VLSB the whole square into d VLSB.
If we solve this equation, we get VLSB divided by root 12.
The equation for signal-to-noise ratio is 2 raise to n VLSB.
Instead of VIN, we can write 2 raise to n VLSB divided by 2 root 2 whole divided by noise.
That is quantization noise.
The equation for signal-to-noise ratio is 2 raise to n VLSB divided by 2 root 2 whole divided by quantization error into quantization error of RMS.
The next property is aperture error.
Aperture error is the maximum sampling error associated with aperture uncertainty can be no larger than 1 by 2 LSB.

