THD Simulation

Hi all,

what is the best way to compute the THD of a circuit (using spectre)?

I could use tran analysis.
- One problem is identify the temporal window time over which perform
the the DFT, i.e. I have to wait enough for the signal's transient to
be extinguished. Maybe doing it manually can lead to some bad
estimations.
- How do I let spectre perform waveform sampling that is good enough
for the THD calculation. Automatic sampling is ok?
- I read the thd() SKILL function uses the dft() function. Does dft()
performs a resampling?

I thought Periodic Distortion Analysis (pdisto) could also do the job
of, at least identifying the steady-state response, if the states are
actually set to be the same one. Is this true? (I never used Periodic
Steady State analysis so I may be completley out of track here).

Thanks.

Marco

 

[Riad KACED]

Hi Marco,

I did some THD calculations when simulating ADCs a little while back.
I have used Spectre TRAN simulation for my case. I have been looking
at other alternatives at the time and I did find that SpectreRF does
support THD with PSS. I didn't use PSS because of lack of experience
at that period. Anyway, you have to be very careful with your
simulation options if you want to get meaningful THD results. The
number of samples is very important, it should be high enough to take
into account all the harmonics. You have to set it at least at twice
the nyquist frequency of your signal. The calculator default is 64 off
the top of my head. You might need to set it at a higher value,
depends on on the frequency information held by your signal.
There is one thing I won't like with the calculator THD function: You
can't specify any windowing. This is really important when your signal
is not periodic or you do not know the period. You can do this with
the DFT function though. The windowing can completely change the DFT
results. The other bit the was annoying me with the tran simulations
is the fact you have to run the simulation for at least 10 periods
(when periodic). The Spectre/Spice book talks about 10%-15% of the
tran time to complete if good enough to allow the time step to
stabilize.

Anyway, If you do follow all the options as in the calc doc, you are
most likely to get a descent THD.

You could try the PSS THD otherwise but I can't make any comment on
this since I haven't used it. Andrew may lighten us with his
knowledge.

BTW, are using any Matlab along with MMSIM ?

Cheers,
Riad.

 

[sai dhiraj]

On Sep 25, 4:26 am, Riad KACED <riad.ka...@gmail.com> wrote:

Hi Marco,

I did some THD calculations when simulating ADCs a little while back.
I have used Spectre TRAN simulation for my case. I have been looking
at other alternatives at the time and I did find that SpectreRF does
support THD with PSS. I didn't use PSS because of lack of experience
at that period. Anyway, you have to be very careful with your
simulation options if you want to get meaningful THD results. The
number of samples is very important, it should be high enough to take
into account all the harmonics. You have to set it at least at twice
the nyquist frequency of your signal. The calculator default is 64 off
the top of my head. You might need to set it at a higher value,
depends on on the frequency information held by your signal.
There is one thing I won't like with the calculator THD function: You
can't specify any windowing. This is really important when your signal
is not periodic or you do not know the period. You can do this with
the DFT function though. The windowing can completely change the DFT
results. The other bit the was annoying me with the tran simulations
is the fact you have to run the simulation for at least 10 periods
(when periodic). The Spectre/Spice book talks about 10%-15% of the
tran time to complete if good enough to allow the time step to
stabilize.

Anyway, If you do follow all the options as in the calc doc, you are
most likely to get a descent THD.

You could try the PSS THD otherwise but I can't make any comment on
this since I haven't used it. Andrew may lighten us with his
knowledge.

BTW, are using any Matlab along with MMSIM ?

Cheers,
Riad.

hai
u guys discussed abt THD calc
can you please elaborate on the method to calculate THD after getting
DFT of the signal?

 

[Riad KACED]

Hi Sai,

THD is the measure of the amount of energy in the harmonics relative
to the energy in the fundamental.
So from your DFT, choose the first N harmonics of interest, let say 20
but it depends on your application, sum the energy of those harmonics
and devide it by the energy of the fundamental.
A more detailed explanation could be found in the Cadence's Waveform
Calculator User Guide, just look for THD. Any other descent Analog/RF
primer explains this as well.

Cheers,
Riad.

 

[mm77]

On Oct 13, 12:45 am, Riad KACED <riad.ka...@gmail.com> wrote:

Hi Sai,

THD is the measure of the amount of energy in the harmonics relative
to the energy in the fundamental.
So from your DFT, choose the first N harmonics of interest, let say 20
but it depends on your application, sum the energy of those harmonics
and devide it by the energy of the fundamental.
A more detailed explanation could be found in the Cadence's Waveform
Calculator User Guide, just look for THD. Any other descent Analog/RF
primer explains this as well.

Cheers,
Riad.

Few doubts:

Suppose the frequency points of the DFT do not correspond to the
harmonics of your signals, which happens when the time window of the
signal over which you applied the DFT is not equal to the period of
your signal (the fundamental). What to do in this case? How to
detemine correctly the power of the harmonics?
Do you now have to intepolate the DFT to get the peaks?

Things are easier if the sampled period *is* equal to the period of
your signal?
I think this can be done with PSS analysis....

How to take noise into account in the calculation? (white noise or
other any other kind)

With added noise, a periodic signal is not periodic anymore... so how
can PSS still succeed in finding a period?

Kind regards,
Marco

 

[Riad KACED]

Hi Marco,
I do apologize for the late answer. Have been very busy over the past
couple of weeks and trying to catch up ...
These are few few answers :

Suppose the frequency points  of the DFT do not correspond to the
harmonics of your signals, which happens when the time window of the
signal over which you applied the DFT is not equal to the period of
your signal (the fundamental). What to do in this case? How to
detemine correctly the power of the harmonics?
Do you now have to intepolate the DFT to get the peaks?
Things are easier if the sampled period *is* equal to the period of
your signal?

There is something fundamental to know here. The Fourier analysis
inherently assumes T-periodic waveforms. if the period of the signal
and the Fourier interval do not match, then a discontinuity will be
generated and this will contaminate your results. In this case, it is
worth to either running a tran to determine the exact period before
running the DFT or apply a windowing function. Spectre assumes a
rectangular window. There is a good explanation of this on the
Cadence's SourceLink. Please give a look at:
http://sourcelink.cadence.com/docs/db/kdb/2006/June/11251190.html
Chapter 5 from the Designer's Guide to Spectre book is very good as
well.
Generally speaking, you have to make a good transient waveform right
from the beginning if you want to end up with meaningful DFT/THD
results. So you need to think about some options like the 'reltol',
the maximum time step, the integration method, the strobing ... etc.

I think this can be done with PSS analysis....

I'm afraid not, Periodicity is one of the fundamental assumptions for
PSS analysis as well.

How to take noise into account in the calculation? (white noise or
other any other kind)
With added noise, a periodic signal is not periodic anymore... so how
can PSS still succeed in finding a period?

Well, noise in all the sorts, i.e 1/f, white, simulation floor noise,
shouldn't be a problem for PSS. This is how PSS works (assuming
Shooting method, you could look at the Harmonic Balance):
PSS tries to find and initial condition that directly results in a
steady state, i.e the node voltages a periodic with the same period.
The simulator compares the solution between the start and the end of
the period. If there is a difference, then it will iterate by changing
the initial V/I point. PSS uses the circuit equations matrix to find
out the efficient way to recompute the initial point. There are more
details about the PSS algorithm in the SpectreRF Theory doc.
Please bear in mind that PSS is a 'complicated' analysis that you need
to tweak for your type of design. In other words, the PSS settings are
different for a VCO, PLL, Mixer, LNA ... etc.

I don't know whether I have covered all your questions or not. I'm
actually not an expert of this topic and hope Andrew could find some
time to help in case of anything missing. Otherwise, I'm usually using
the Designer's Guide Forum for this kind of topics. There are loads of
threads about the FFT, THD, PSS over there and it's worth sparing some
time.

Regards,
Riad.