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.

104 MMSE AND ML DECISION FEEDBACK EQUALIZATION<br />

which implies<br />

h = wl/wi. (5.27)<br />

Now that we have X2, we can subtract the influence of si used the detected value.<br />

However, x\ now has a copy of s^ as well. So, now we need to also subtract the<br />

influence of s\ on x\, We can then use the result to detect «2 with a weighted sum:<br />

where<br />

«2 = viwVw + W42/4, (5.28)<br />

2/3 = xi - (wic + w 2 e)si (5.29)<br />

2/4 — %2 — ( e + hc)s\. (5.30)<br />

This looks like another MMSE linear equalization design problem. Thus, we can<br />

use MMSE linear equalization design, with models for y :i and 2/4, to determine good<br />

values for w-.t, and w\ (see the Problems).<br />

5.3 THE MATH<br />

Similar to Chapter 4, MMSE and ML formulations are given. Other design criteria<br />

are not discussed, as they would be the same as in the previous chapter.<br />

Performance results are also provided.<br />

5.3.1 MMSE solution<br />

With MMSE DFE, the received signal is initially processed by a partial MF, producing<br />

received sample vectors. These sample vectors are processed by a forward<br />

filter, which collects signal energy and suppresses ISI from future symbol periods.<br />

The FBF removes ISI from past symbol periods. Unlike the chapter on ZF DFE,<br />

we allow for arbitrary pulse shaping, fractionally spaced sampling, and arbitrary<br />

path delays.<br />

The design of the MMSE FBF is similar to the design of the ZF FBF. Detected<br />

symbols are modulated and channel filtered and then subtracted from the received<br />

samples. Design of the MMSE forward filter is similar to the design of the MMSE<br />

linear equalizer. The only difference is that the FF works on modified received<br />

samples, modified to remove ISI from past symbol blocks. This simply changes<br />

the computation of the data correlation matrix R. Compared to the ZF FF, the<br />

MMSE FF works on future samples in addition to the current sample. Like matched<br />

filtering, this allows better collection of symbol energy.<br />

Assuming partial matched filtering at the front end, recall from (1.23) and (2.59)<br />

that the received samples can be modeled as<br />

where<br />

oc<br />

v{qT s ) |= y/W, Σ M« 7 '» - m7 ')s(m) + n{qT s ), (5.31)<br />

m= — 00<br />

£-1<br />

h(t) = ^gtRpit-Ti). (5.32)

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

Saved successfully!

Ooh no, something went wrong!