19.11.2014 Views

mohatta2015.pdf

signal processing from power amplifier operation control point of view

signal processing from power amplifier operation control point of view

SHOW MORE
SHOW LESS

Create successful ePaper yourself

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

174 PRACTICAL CONSIDERATIONS<br />

In practice, we have to estimate these quantities or related quantities. As these<br />

parameters change with time, we need to adapt the equalizer over time, giving rise<br />

to adaptive equalization.<br />

To understand what options we have, let's assume we are building an MMSE<br />

linear equalizer. One option is to estimate the channel response and noise power<br />

and use them to compute the weights. We refer to such an approach as indirect<br />

adaptation of the equalizer, because we adaptively estimate one set of parameters<br />

(channel response and noise power) and then use them to calculate another set of<br />

parameters (equalization weights).<br />

How would we estimate the channel response and noise power? It depends on the<br />

particular way signals are transmitted. Often known symbols are sent, referred to as<br />

pilot symbols. These symbols may be clustered together into a synchronization word<br />

or training sequence, which can occur at the beginning of set of data (preamble) or<br />

in the middle of the data (midamble). The receiver searches for this known pattern.<br />

Finding it gives us the packet or frame timing. We can then estimate the channel<br />

response as follows. Suppose the known symbols are s«, si and s 2 and suppose we<br />

have determined that we only need to estimate two channel coefficients, c and d.<br />

We can estimate these using<br />

c = (0.5)( Sl r 1 +s 2 r 2 ) (8.1)<br />

d = (0.5)(sori+sir 2 ). (8.2)<br />

These operations are correlations (weighted summations), as they correlate the<br />

received samples to the known symbol values. Why does it work? Let's substitute<br />

the models for r\ and r 2 and see what happens to c. Using the fact that s^ = 1,<br />

we obtain<br />

c = (0.5)[«i(csi+dsii+ ni) + s 2 (cs 2 +dsi+n 2 )] (8.3)<br />

= c + 0.5d(siSo + s 2 si) + 0.5(«ini + s 2 n 2 ). (8.4)<br />

The first term on the right is what we want, the true value c. The second term<br />

is interference from the delayed path. It would be nice if this were zero. Because<br />

we are transmitting known symbol values, we can select their values such that<br />

siso + s 2 «i is zero! The third term is a noise term which has power (1/2)σ 2 . This<br />

is half the noise power of a single received sample because we have used two received<br />

values to estimate c.<br />

In general, if we have N s +1 known symbols and use N s received values, the noise<br />

on the channel estimate has power (1/N s )a 2 . Thus, while the channel estimate is<br />

noisy, it can be made much less noisy than the received signal.<br />

Once we have channel estimates, we can obtain a noise power estimate by subtracting<br />

an estimate of the signal and estimating power. Specifically, we have<br />

σ 2 = (0.5)[(n - cs! - ds () ) 2 + (r 2 - cs 2 - ¿ Sl ) 2 }. (8.5)<br />

Thus, with indirect adaptation, we estimate the channel response and noise power<br />

from the received samples and use these estimates to form the equalization weights.

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

Saved successfully!

Ooh no, something went wrong!