harmonic distortion in S/H circuit

[jiangxb]

hi all

i'm designing a S/H circuit using K. Haug's narrowband-compensated SC amplifier. i want to attain 120dB SNR, but the third harmonic distortion is very large, even though using ideal opamp, i don't know why.
input signal 5kHz sinewave, sampling frequency 1.6MHz

 

[Btrend]

firstly, u can check that with ideal switch
seconldy, refer to this

 

[jiangxb]

hi btrend

i've found some strange cases: when using ideal switch the odd harmonic distortions are larger, and when using ideal op amp the odd harmonic distortions are larger too, but when using real switch and op amp together and adjusting properly the odd harmonic distortions are decreased. maybe the nonlinearity of switch and opamp gain cancel each other, i think.

 

[watersky]

I think you should use ideal switch and ideal amp. If result is good, you can displace one of them. You can find which is primary effection. If result is worst, perhaps clock is reason.

 

[hr_rezaee]

Hi
Use better input switch such as bootstrap switch.
use bottom plate sampling.
Spice can not simulate ideal elements good.
use more accurate simulation. (reduce step time)
regards

 

[jiangxb]

hi watersky

i try to simulate this circuit using ideal opamp and ideal switch, but it isn't convergent. how should i do?


hi hr_rezaee

i've used bottom-plane sampling.
using boosted switch has reliability question, i think, moreover, i've use ideal switch but it has no effect.
using smaller step time and smaller tolerance has no effect either.

 

[analogy]

Hi,


This problem may be related with the choosen frequency values. I suggest you to do coherent sampling by applying this formula fin=fsampling*Nperiod/Npoints.
N points must be a power of two.

I hope it helps.

 

[jerryzhao]

1.6M over 5K =320. 320 is integer. So harmonic large even you use window function. U can select 5.01K input, and try it again.

 

[jiangxb]

hi analogy and jerryzhao

i make new simulations following your methods, but it isn't no effect yet.

thank you!

 

[bbbb]

May be its problem of the simulator (is it in high-accuracy mode?)

are the rise/fall tiems short enough?

when sampling fom a DC input is output accurate?

 

[jiangxb]

hi bbbb

i ever used high resolution mode, but the result is same.
the rise/fall time has no effect on result, and the longer the rise/fall time is, the better, i think.
when sampling a DC input, the output is accuracy.

thank you anyway.

 

[jerryzhao]

Witch tool and window function you used to do FFT.
As the samples are finite, so we need window function. and should better don't let sample rate is integer multiple of input signal's frequency. If you have do that, I think you should check your SNR program.

 

[jiangxb]

hi jerryzhao

i use matlab to process the data. the window function is hanning: w=0.5*(1-cos(2*pi*(0:npts-1)/npts)), and some important instructions as blew

aout=in((np-npts+1):np).*w';% np is total points
y=fft(aout);
psdy=(abs).^2/npts;
plot(10*log10(psdy(1:200)));

 

[chungming]

jiangxb:
you can try this

Fin = 3.3203125K Hz
simulation = 8.2m s
FFT 8192 points from 3.011764706e-3 to 8.131764706e-3 s

 

[jerryzhao]

Compute your samples number. let them is whole Sin signal. you can change your sample's points and do fft. you know if your sample points is not whole signal cycle, that will contribute the harmonic too.

 

[jiangxb]

hi chungming

i have followed analogy's advice to input 3.90625kHz, and sample 10 cycles(4096 points), but the results is same.

hi jerryzhao

yes, i processed the whole cycle. when i change the cycle, the same. while change the number of sample point, it will have spectral leakage.

 

[jerryzhao]

1 Did you set your Vos=0V?
2 Check your input is whether overflow.
3 Check your capacitor retio. (check your gain whether right.)

Added after 10 minutes:

1 Did you set your Vos=0V?
2 Check your input is whether overflow.
3 Check your capacitor ratio. (check your gain whether right.)

 

[jiangxb]

hi jerryzhao

i use full-differential topology, Vos=0V, input and output signal don't overflow, and gain=1. i think these are normal.

 

[jerryzhao]

system gain=1???Why your signal power peak vaule is 40dB in fft frequency spectral.(figure untitled.jpg)

 

[jiangxb]

hi jerryzhao

yes, i'm confused by this question.

i use this function:

aout=out((np-npts+1):np).*w';% np is total points, npts is the points processed, out is the output of PGA
y=fft(aout);
psdy=(abs).^2/npts;
plot(10*log10(psdy(1:200)));

i don't know why the spectral density isn't corresponding to real case. where is wrong?