Chungming,

  There are currently no standard calculator functions for SNR
and SINAD. You will need to write your own calculator functions
to implement the functions. One other option would be to dump
the FFT magnitudes to Excel and perform the calculation there.

Calculating SINAD is easy you just
need to sum up all the FFT bins in the band of interest. Calculating
SNR and THD are more complex because you need to account for
frequency folding due to sampling, that is the distortion tones
are aliased back into the Nyquist bandwidth need to be correctly
accounted for in the calculation.

Chungming,

  SNR is Signal to Noise Ratio and does not include distortion
  SINAD is Signal to Noise and Distortion Ratio, also called, SNDR

  You will need to perform the calculations in the power domain
   or use RSS in the voltage domain.  Measure the power in the
   Fundamental, at the input frequency, and the total power.

  SINAD = 10 log | (Fundamental) / ( Total -Fundamental)|

  Use SINAD when calculating effective number of bits

  SNR is more difficult since you need to account for the distortion
  tones, 2*fin, 3*fin, ...

   SNR  = 10 log | ( Fundamental) / ( Total - Fundamental - Distortion)|

   Normally transient analysis only includes numerical noise and shaped
quantitazation noise. You will need to use Spectre's transient noise analysis
to include the effect of device noise on the SNR. You will also need to
create a "noisy" clock, use Verilog-A, to model the effect of clock jitter on
SNR and SINAD.  

Also, there is one other option, you can calculate the SINAD directly
using the Spectre's Fourier Analysis.  It is very accuracte but has two
limitations,
1) It is uses the Rectangular Window function, for a Sigma Delta
   you should use a windowing function. You can use Verilog-A
   to create your window function so this is not a big issue.
2) The analogLib implementation of the four component is incomplete
    so you will either need to run from the command line or edit  the
    CDF to add complete syntax to the component.

                                                                   Sheldon

I use the fourier and fourier2ch components in the analogLib to calculate the SFDR and THD. The component will automatically invoke a FOURIER analysis and deliver the nonlinearity in term of THD. You only need to simulate one period of the sinusoid signal. The result is more accurate than the conventional method with FFT.

To automatically extract the SNR and SNDR, you can use OCEAN function of Cadence. It is actually very easy to program.

One comment,

  Gita is correct about the Fourier component, however, the Fourier
component does not support windowing. If you simulate a Delta Sigma
Converter and analyze the results with the Fourier, you will need to
include a window function in the simulation. You can use Verilog A
to create a window function behavioral model, then the model to
filter the Converter output before the Fourier component.

                                                                         Sheldon

Ali,

 You might try using log for the x axis, that will make the results
look a little better. However, there is a large tone everywhere. Not
sure about the source, low order delta-sigma with dc input?

                                                                  Sheldon

Ali,

Some comments:

1) If you plot on a logarithmic scale the results will look better

2) Your circuit has a problem at low frequencies the transfer function
    continue to decrease as frequency decreases. In your case the
    transfer function flattens out. So there is a problem with your
    design that is preventing you from achieving the result you
    are expecting to see.

3) Is your sampling frequency 10Hz or 2.56kHz? Or why do you
   need the external S/H in this case? Since your trying to debug
   a problem, why don't you use a fixed input and skip the
   extra S/H to avoid the spikes everywhere, it will make it
   easier to identify and correct the issue.

4) 64 points * OSR = 64 * 256 = 2^6 * 2^8 = 2^14 = 16384
   in this case there are other problems so this is not really
   relevant