time derivative of discrete data in simulink

I'm implementing a program in Java that was delivered in Simulink. My expertise is limited, and I'm stuck on converting a derivative block.

The simulink code applies a du/dt block to the input data. I'm doing the following in my code:

du_dt = (signal[i] - signal[i-1])/time[i];


where signal is an array of data samples, and time is an array of elapsed time (in seconds) between sample i and sample i-1.

my java program is generating much larger derivative values than my simulink program. doing some digging, i see that the simulink man8al talks about the derivative definition as

$$y(k)=\frac{1}{\Delta t}(u(k)-u(k-1))$$

and taking the z-transform

$$\frac{Y(z)}{u(z)}=\frac{1-z^{-1}}{\Delta t}=\frac{z-1}{\Delta t\cdot z}$$

my undergrad in mathematics stops short of being able to understand what the z-transform is and how to translate it into code. any help?

EDIT

EDIT 2

As requested, here's the data of the Simulink output and mine:

0.0, 0.0
1211.4359554191567, 692.2491173823755
573.0790390610672, 859.6185585916019
346.12455869118764, 461.4994115882501
807.6239702794378, 634.5616909338439
230.74970579412505, 692.2491173823755
286.53951953053365, 401.1553273427467
519.1868380367816, 634.5616909338443
-288.43713224265656, 403.8119851397189
-57.307903906106624, 343.8474234366401
-173.06227934559388, 57.68742644853153

0   0
1.01133071339275    1.01133071339275
-1.38227619434410   -1.38227619434408
-1.43825420168704   -1.43825420168706
-0.640385232441365  -0.640385232441365
-0.157483736804270  -0.157483736804270
-1.01133071339275   -1.01133071339275
-2.92203612549265   -1.96668341944271
-3.77588310208105   1.68679338272849
-2.98846441805972   1.44870448691124
-2.96756384761120   -0.629934947217123


The first section is mine, the second is the simulink. The indices are aligned, the data is two-dimensional but the derivative is of each individual dimension (so it's like two separate data sets)

-
What does i, signal[i], and time[i] represent? Are they scalar? Of what type (integer, rational, complex, etc)? And are they discrete or continuous? – user2468 Apr 1 '12 at 23:39
i is an integer index into the signal and time arrays; signal is an array of discrete signal samples, rational real numbers; time is an array of time elapsed between sample[i] and sample [i-1], in seconds. – kolosy Apr 1 '12 at 23:53
– user2468 Apr 1 '12 at 23:57
yes, i saw that before i came here. theoretically relevant, but definitely not what the simulink documentation is talking about – kolosy Apr 1 '12 at 23:58
Your arrays are in time domain. You do not need the discrete Laplace transform. The simulink page describes the discrete time derivative as (signal[i] - signal[i-1])/time[i]. Also, it describes the derivative in $z$-domain as (singal[z] - signal[z]/z)/time[z]. Your arrays are in time domain. – user2468 Apr 2 '12 at 0:03

Regarding your question about implementing the z transform in code: The z transform is a one time frame advance in the sample series. The 1/z transform is a one time frame delay in the sample series. So you can see directly that your java code implements ${1 - z^{-1}\over \Delta t} = {z-1\over z \Delta t}$. The ${1\over z}$ transform is easy to implement in computer code. Just retain the signal value from the prior time frame. The only problem is in deciding how to initialize the prior frame value for the first call. On the first frame, it is normally best to initialize first-frame prior values as though the signal has been steady. That avoids initial transients that may take a long time to die out. So your Java code would start with signal[0] = signal[1]; More complicated Tustin transformations that use z would require solving for the correct signal prior values that the first frame should use for a steady state initialization.

-

The simple command

du_dt = (signal[i] - signal[i-1])/time[i];

is a correct translation of time derivative. One should be aware, though, that this approach greatly amplifies noise. Several noise-robust differentiators are described by Pavel Holoborodko.

Looking at the output of Java, I conclude that the input was not the same in both columns. Yet, the simulink output is the same in both columns most of the time. This suggests that two programs do not really do the same thing. There is not enough information to find exactly what the second one is doing.

But there is something notable in rows 5-6 of java output:

807.6239702794378, 634.5616909338439
230.74970579412505, 692.2491173823755


Namely,

807.6239702794378 - 230.74970579412505 = 576.87426448531275
634.5616909338439 - 692.2491173823755 = -57.6874264485316


There may be some sign and order-of-magnitude issues here.

-