Browse Prior Art Database

Efficient Implementation of Integrated Branch Metric Unit for Equalizer and Convolution Decoder

IP.com Disclosure Number: IPCOM000020063D
Original Publication Date: 2003-Oct-22
Included in the Prior Art Database: 2003-Oct-22
Document File: 5 page(s) / 90K

Publishing Venue

Motorola

Related People

Mohit K. Prasad: AUTHOR [+5]

Abstract

An efficient realization of a common circuitry for computing the branch metrics of a GMSK equalizer and a Viterbi decoder such that the rest of the hardware (ACS and trace-back units) can be shared.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 50% of the total text.

Efficient Implementation of Integrated Branch Metric Unit for Equalizer and Convolution Decoder

Mohit K. Prasad, Arnab K. Mitra, Nitin Vig, Gaurav Davra, Amritpal Singh

Abstract

An efficient realization of a common circuitry for computing the branch metrics of a GMSK equalizer and a Viterbi decoder such that the rest of the hardware (ACS and trace-back units) can be shared.

1.    Introduction

In a GMSK receiver for GERAN systems the algorithm for GMSK equalization and convolutional decoding both use the Viterbi algorithm. It therefore makes sense to share the hardware for the two as is customarily done in most GSM designs. The add-compare-select unit (ACSU) and the trace-back unit (TBU) can be shared without significant changes. The key difference lies in the manner in which the branch metrics are computed for each of these two cases. The difference in the branch metric unit (BMU) becomes more pronounced when the EDGE functionality is added. EDGE supports a much lower rate code (upto rate 1/10) as compared to rate 1/6 of GSM. This paper addresses the problem of efficient implementation of the combined branch metric unit. The same method could be extended for BMU for other equalizer and channel decoder combined units.

For channel equalisation, the branch metric for each transition within a Viterbi butterfly is a func­tion of the Matched Filter output (MF) and the L-Metric Viterbi Parameters (VP) as shown in Figure 1. The VP values come from the channel sounding. After extracting the channel impulse response coefficients via a cross correlation process (also referred to as S-Parameters and used as the ISI FIR coefficients), the VP value for a particular state ‘wxyz’ is usually calculated as:


Equation no. 1 

The VP values calculation is done by software within the DSP core. The VP values have symmet­rical property,


Equation no. 2 

For convolutional decoding at the receiver, the received symbols are quantized and soft symbols,

are then used to compute the branch metric values, where

is the code rate and

. For any hypothesized codeword output,

,

the branch metric is a measure of the transition probability from the hypothesized codeword to the received codeword as the encoder transitions from state

to state

(

) and is given by

Equation no. 3

2.   Description

A system which could calculate the branch metrics for both the equalizer and decoder would require the following aparatus:

1.      Interface which would take as inputs the soft symbols required during decoding and S-parame­ters required during equalisation. The soft symbols are required to be fed for each trellis stage whereas the S-parameters are constant within a frame. The matched filter output (MF) is required during equalisation for each trellis stage but MF is added to the selected path metric in Add Com­pare Select (ACS) unit. The technique of using MF in the ACS may be derived from Figure 1.

2.      Logic to multiplex the soft symbols and S-parameters. This would require a multiplexer which sele...