22.01.2014 Views

download searchable PDF of Circuit Design book - IEEE Global ...

download searchable PDF of Circuit Design book - IEEE Global ...

download searchable PDF of Circuit Design book - IEEE Global ...

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Network Objective Functions 151<br />

and frequency. The second integrand might represent the difference between a<br />

response function (R) and the goal function (G).<br />

Since integration on digital computers is discrete anyhow, the measure <strong>of</strong><br />

goodness <strong>of</strong> fit can be a process <strong>of</strong> frequency sampling. The Euclidean norm<br />

(inner product) mentioned in (5.82) applies here as well:<br />

2 2 1/2<br />

IIEII=(el+e~+.. ·+eN) . (5.87)<br />

This might correspond to sampling at the ith frequency, where ej is the<br />

difference between the response and the goal. The next section combines these<br />

concepts in a form convenient for optimizing network response functions<br />

sampled at several frequencies or times.<br />

5.5.2. Discrete Objective Functions. A typical discrete objective function for<br />

network response is shown in Figure 5.25, as described mathematically by<br />

M<br />

E(x,w)= 2: W;(R,-Gj)P, (5.88)<br />

i=l<br />

where P is an even integer (the Pth difference), R, is the response, G j is the<br />

goal, and Wj is the weight factor at the ith frequency. None <strong>of</strong> these quantities<br />

are complex. For example, if a network is to be adjusted so that an impedance<br />

approximates some given impedance values at various frequencies, then an<br />

approximate response might be SWR, according to (4.59) and (4.54). Compare<br />

(5.88), with P=2 and Wj= I, to (5.87). Also, (5.88) may be generalized to<br />

account for more than one kind <strong>of</strong> response, R;k' by adding a second, nested<br />

summation on k. Two responses might then be'SWR and voltage, where the<br />

weights W, k must equalize the scales for the two different kinds <strong>of</strong> responses.<br />

In practic~, only very few kinds <strong>of</strong> responses are successfully considered<br />

simultaneously, and there is a good chance for a stand<strong>of</strong>f (over constraint), so<br />

that optimization is ineffective.<br />

A "satisfied-when-exceeded" feature can be included in a program for<br />

(5.88), so that W,=O is employed whenever R,>G,. This feature is useful<br />

Least Pth:<br />

Response<br />

Figure 5.25.<br />

Least·Pth error function with weighted frequency samples.<br />

w

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!