You could do that (make sure that you set strobeperiod on the transient analysis to coincide with your sampling instant to remove the interpolation error in the dft sampling).

Or you could use PSS or QPSS to get this directly, I guess.

Regards,

Andrew.

ericjohnson,

  To strobe the output use the Spectre tran command,
skipstart, to define the point where you would like to
start the data record for the FFT and use, strobeperiod,
to set the fixed time for the FFT. For example, assuming
you have a 10MHz clock and would like to allow 10 periods
for the start-up transients to settle,
tranAna tran ....   skipstart=1.098u strobeperiod=100n
Remeber: that you allowed 10 periods to settle transients,
1u, and you need the S/H to acquire, switch into hold mode,
and settle, 98ns.

  The other options are to use the zvcvs components
and or the zvcvs component with the fourier component.
The zvcvs acts like an ideal sample and hold. By
performing the FFT on the output of the zvcvs, you can
get very accurate results. BTW, I use the zvcvs. The
setup of the zvcvs is similar to the setup of the strobe.

                                                      Best Regards,

                                                          Sheldon

Ericjohnson,

  First, don't use 10MHz, selecting the input frequency
requires a lot of care. Next, the number of samples needs
to be a power of 2, if you use standard FFTs.  So to setup
your simulation, assume 128 point FFT and use input
frequency of 10.156250MHz( 13/128 ), 10 clock cycles
to settle the start up transients, 100ns, and that the
output is valid at 9.8ns into the clock cycle.

tranAna tran stop=1.4u skipstart=109.8n strobeperiod=10n

In the calculator, after selecting the waveform, vt -->
V(out)[exact syntax may be different]
Then select dft
from 109.8n to 1.3898u number of points 128
you can use the rectangular window. Don't forget
to convert to dB20 before plotting.

The alternative, you can sample the output node with
a zvcvs source

sampler zvcvs sampled 0 out 0  td=109.8n ts=10n

and then
1) perform an FFT on the output, see above
2)use Spectre's fourier analysis
   
fourAna sampled 0  fourier fund=10.156250M harms=64
normharms=13 where=logfile order=2


                                                       Best Regards,

                                                           Sheldon

Eric,

  First, comment we are simulating not measuring
performance on a test bench so we are taking advantage
of the simulator's "ideality" to make the analysis better.

  Some general guidelines for a well-behaved FFT of
s/h and ADCs are(See Ken's book for more details):

1) The data record for the FFT should include an integer
    number of evenly spaced(in time) data points.
     --> This can be a issue, which is one of the reasons
            why Spectre uses the Fourier integral

2) The data record should have an integer number of
    cycles of the input and the sample periods.
    --> This avoids needing to use window functions.,
           that is, there is no spectral leakage so windowing
           is not required.

3) The ratio of the input frequency to the sample frequency
   should be a prime number.
   --> For ADCs in particular, if the input frequency is a
         related to the sample frequency, fin=10MHz and
         fs=100MHz. Then only certain codes get exercised
         and the spectrum has discrete tones at the
         harmonic frequencies: 10MHz, 20MHz, ..
         While the SNR calculation is correct, the SFDR and
         THD calculations are not. All the quantatization
         noise ends up in the harmonics.

4) The 10 periods is a WAG. Transient analysis
    takes a while to settle the startup transient. If
    you think about PSS, the shooting algorithm eliminates
    the transient start-up and directly calculates the
    steady-state. In this case, we allowed 10 periods for
    the circuit to reach "sinusiodal steady-state". In
    complex designs, circuits can sometimes require long
    times to settle startup transients.

5) The 9.8ns is a WAG. Assuming that 5ns is allowed for
    sampling and 5ns are allowed to settle the hold step,
     then the hold step should be "settled" at about
     4.8ns after the start of the hold period. If you wait
     to long, then the S/H starts to switch back into
     sample mode and the result is not valid. Since
     simulators tend to anticipate transistions, you need
     to allow a little margin.

6) Given your targets of fin=10MHz and fs=100MHz, and
    your (anticipated) desire for a "reasonable" simulation
    time. The selections were:
    a) 128 points for the FFT, power of 2, with reasonable
         accuracy.
         Increasing the number of points doubles the
         simulation and lowers the noise floor. Increasing
         number of points in the FFT reduces the
         "resolution bandwidth" of the bin, i.e., less rt Hz,
         less bandwidth less noise
     b) Your target input frequency is 10MHz so 13 is a
          prime number that results in an input tone that
          is close to your target.


  Hope this helps!

                                                          Best Regards,

                                                             Sheldon

sheldon, I was confused by the following questions:

1. After checking the clock plot, I noticed the clock holds at either high or low (i.e., not a pulse train anymore). Why is that? And why the SHA outputs follow the input sine wave with a little distortion/delay? I mean, since after the initial 10 clock cycles, the clock stays at high, then the outputs should hold at the last value the SHA sampled, right?

2. Why for other circuits, we don't wait for a while when using the transient analysis?

Thanks again!!!

Tons of thanks...

But here comes the question again: from your description, this strobing function is sort of like an impulse sampling. However, we are trying to measure/simulate a pulse train sampling, which has different spectrum. Then seems to me the results obtained by strobing is not accurate...

First of all, thank sheldon for his patience and insightful answers! Here's a summary.

Q - What's wrong with my original simulation setup?

A - From your initial post, fin=10MHz and fs=100MHz, with simulation time of 1005s.

1) So the FFT has 100 data points, and this is not a power of 2.

2) The input and clock frequency are harmonically related, that is the clock is the 10th harmonic of the input, or only input states exist, so the quantitazation noise will show up in the harmonics of the input

Q - "thd" function in calculator

A -  the THD function is not applicable to this
  problem, i.e., S/H and ADC THD calculation(it works only
  in special cases). The THD is intended for non-sampled
  circuits. In sampled-circuits the harmonics are "folded"
  back into the baseband. So the 6th harmonic appears at
  40MHz, the 7th harmonic at 30MHz, ...

For your current simulation this is not a big issue, but, sorry, you need to do the THD calculation manually.