DSP56300 datasheet(63/108 Pages) FREESCALE

63 / 108 page

Algorithmic Extensions

Main: Putting the Pieces Back Together

Viterbi Decoder Implementation

4-15

To make the changes to the MAIN program code, we begin as we did in Section 3, by

initializing the address registers. R2 is still used to address branch metrics. Registers r4

and r5 address the states for path metric storage. The differences needed for the

initialization code are to set n5 to 1 (needed for tracking the number of states in the

PreAcs code), and n2 is set to 1. This is different from the code in Section 3 for two

reasons. First, the branch metric generation is different, and second, using n2 to update

the addresses to read branch metrics for the ACSFlush routine.

Next, we do ENCBITS-1 iterations of input processing to preload the path storage

registers. This is different because we use the PreACS macro instead of the ACS macro.

Using the PreACS macro means that the code uses the starting encoder 0 state to reduce

execution cycles as well as to avoid the need of initializing the path metric storage for all

the states.

The next major piece of code is the nested do loops that process bytes of decoder input

and store off the resulting paths. This code is identical to the code in Section 3, except

that we process one less byte of data.

The end processing is more complicated because there are two subprocesses occurring at

the same time and their endings are not in phase. One process is the storing of the path

data in bytes as the path storage registers fill up. The second process is the ACSFlush

processing, which collapses the trellis states back to 0.

For this example, start with 168 decoder inputs. We preprocess five inputs, and then

process 19*8=152 more inputs in bytes using the nested loops. We have 11 decoder

inputs left to process. We need to do 8 more to get the next byte, but after processing

6 inputs, we have to change from using the ACS macro to using the ACSFlush macro

(used to collapse states for the last five inputs).

Accordingly, we process the first 6 of the last 11 decoder inputs using the ACS macro.

We then process the next two decoder inputs using the ACSFlush macro. Now we have

processed eight more decoder inputs, and can store off the path data using the

STOREPATHS macro. We then finish processing the last three decoder inputs using the

ACSFlush macro. The final path data is in the collapsed state (as it was in the code in

Section 3

), and is access in the TRACEBACK macro as before. After doing the traceback

using the TRACEBACK macro of Section 3, the decoder is done at the FINISH label.

Freescale Semiconductor, Inc.

For More Information On This Product,

Go to: www.freescale.com

DSP56300 датащи(PDF) 63 Page - Freescale Semiconductor, Inc