ZOH Frequency response and matlab 1

[skhatana]

I am struggling little bit in trying to understand the frequency response of the ZOH with matlab. I was wondering if you have any thoughts on this.

Here are few things that I understand ( coorect me if these are not right)

1) ZOH is linear but not Time Invariant

2) The frequency response of the ZOH is given by H_zoh(s) =( 1- exp(-sT)) / s. This is works for the fundamental component of the ZOH output. Since the actual output of the ZOH consists of lot of high frequency components due to fast transitions.

H_zoh(s) has low pass like characteristics with zero on multiples of the sampling frequency

3) c2d(sys,'zoh',Ts) is a matlab function used to discretize continous domain system (sys( to discrete domain with ZOH and sampling time of Ts.

Problem:

To get the frequency response of ZOH I have set the sys = tf(1,1). When I try the bode of resultant discrete system I get perfectly flat response with frequency ( Basically I don't see H_zoh(s) characteristics).

In past I have explicitly used the explicit expression of H_zoh to do analysis in frequency domain. I was wondering why matlab c2d don't take in account the response of the H_zoh ?


Also,

The output of the ZOH can be approximated as input signal delayed by Ts/2 , Ts being the sampling time. I was wondering when is this approximation valid?


I was wondering if you have good reference on his kind of stuff. I looked into following text w/o getting a coherent description of this issue.

1) Franklin and Powell , Digital Control of Dynamic systems.
2) Ogata, Discrete time Control Systems

 

[sreid]

You might try posting this question at the DSP Blog.

http://www.dsp-bg.info/

 

[IRstuff]

EVen posting in the Matlab forum might make more sense.

ZOH cannot be linear, by definition.

TTFN

 

[skhatana]

Hi TIFN,

Thanks for your pointer to matlab forum.

Actually I did a bit of search on this. ZOH is linear and this aspect is well documented too.

It satisfies superposition and scaling property. It is sruprinsing but true. Try it !

-SK

 

[IRstuff]

It has discontinuities in the first derivative

TTFN

 

[skhatana]

TIFN,

Is discountinuity in first derivative a condition for the system to be Linear ?

I thought following two conditions are suffcient for a systemto be Linear .

1) Superposition : Sum of outputs is same as output of sum of inputs

2) Scaling : Scaling input by a conatct scales output by same factor.

Cheers
-SK

 

[IRstuff]

So you're saying that
ZOH(t₁+t₂) = ZOH(t₁)+ZOH(t₂)

for any t₁ or t₂?

TTFN

 

[skhatana]

Yes.

Just to make sure i have understood your notation. I am restating it.

Superposition:

Given:
y1(t) = ZOH {x1(t)} and
y2(t) = ZOH {x2(t)}

Implies

y(t) = ZOH {x1(t)+x2(t)}= y1(t)+y2(t)


Scaling:

If y1(t) = ZOH {x1(t)}

Then

a *y1(t) = ZOH {a*x1(t)}

 

[IRstuff]

ZOH is a function of sampling time only.

TTFN

 

[jsolar]

In your script you converted sys = tf(1,1) to a discrete system by using c2d(sys,'zoh',Ts). I interpret this as you asking for MATLAB to convert your flat frequency response continuous system sys into a discrete equivalent using the zoh method. This is different from asking for a digital equivalent of a zoh system. You got what you asked for. The zoh equivalent of a flat system is flat, not a zoh system.
sysZohA = tf( 1, 1 );, Ts = 1;, c2d( sysZohA, Ts, 'zoh' )
gives: Transfer function: 1

Since you have stated that the zoh is already known to be H_zoh(s) =( 1- exp(-sT)) / s, you could ask MATLAB to determine the best fit discrete system. We could use a Pade approximation of the delay exp(-sT), then build the zoh(s). [ bPade, aPade ] = pade( Ts, 5 );, sysZohA = tf( [ aPade - bPade ], [ bPade 0 ] )
Transfer function:
-2 s^5 - 840 s^3 - 3.024e004 s
---------------------------------------------------------------
s^6 - 30 s^5 + 420 s^4 - 3360 s^3 + 1.512e004 s^2 - 3.024e004 s
This will get you an approximation to a ZOH in the continuous s domain. Then sysZohD = c2d( sysZohA, Ts, 'zoh' ) should work. I have not chekced to see if the order is backwards, but it should be a discrete approximation using the zoh method of an approximation to a continuous domain approximation of the zoh.
John Solar