Let, the channel be modeled as a FIR model with order L. Mathematically, $y_n = \sum_{k=1}^L h_k u_{n-k} \tag{1}$ where $u$ is the data input to the channel and h is the channel coefficient. In statistics, the model y is known as the Moving Average process with order L. In signal processing, L is known as the length. Consider an example where there are 3 coefficients viz. $h_1,h_2,h_3$

Confusion 1: My confusion is, what is the order for this example? There are 3 coefficients so the order should be 3. I am confused between these terms - order, length and delays.

Confusion 2: Can somebody please clarify what is the order for this example? Please correct me where I have made any mistakes.

In Matlab I have implemented the equation(1) as follows:

N = 1000;
u = rand(1,N);
for n = 3 : N
    y(n) = h(1)*u(n) + h(2)*u(n-1) + h(3)*u(n-2);

In your example order is $2$ and length is $3$.

Its has more to do with convention. Consider a polynomial in $x$: $$ a_n x^n + a_{n-1} x^{n-1} + ... a_0 x^0 $$ The order of this polynomial is $n$, whereas, number of coefficients is $n+1$. In your case an FIR filter of order $L$, is a polynomial you write as: $$ h_L \delta(n-L) + h_{L-1} \delta(n-L+1) + ... h_0 \delta(n)$$ The $z$-transform of which is: $$ h_L z^{-L} + h_{L-1} z^{-L+1} + ...+ h_1 z^{-1} + h_0 z^{0}$$ Note this transform, which you might not have studied yet, just places the shift in $\delta(n)$ on power of $z$, e.g., $z^{0}$ and $\delta(n)$ correspond to no-delay, just scaling the input signal. $z^{-1}$ and $\delta(n-1)$ correspond to order $1$ and one delay.

I think real confusion comes from MATLAB, as it does not allow $0$ as first index value. You cannot write x(0) in MATLAB. The first index is $1$. Therefore, we have to start from x(1).

  • $\begingroup$ Thank you for this answer. Can you please suggest an article or tutorial for understanding about the z transform and their relation with filters, without going into much depth. I am learning estimation theory and system identification. When reading research articles, I am coming across z-transform and unable to understand them as I don't know what is their connection to these filters. These are not there in time series models, of which MA and AR models are popularly studied. $\endgroup$ – SKM Aug 21 '17 at 6:07
  • $\begingroup$ Only in signal processing I have come across z transform assiciated with these time series models. As a result of which I cannot understand why AR filter is a truncated FIR filter. $\endgroup$ – SKM Aug 21 '17 at 6:07
  • 1
    $\begingroup$ If you are familiar with Laplace transform, you can start with this book, here is link to chapter 33: dspguide.com/CH33.PDF $\endgroup$ – learner Aug 21 '17 at 6:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.