gsm_viterbi_soft detection.txt

在GSM通信系统中

在GSM通信系统中，Viterbi检测是一个及其重要的部分，许多的研究人员在这里花费了大量的心血。参考各种资料，我们总结得出以下Viterbi软输出算法，与传统的硬判决方法相比，可以提高2～3个db的增益：



function [ rx_burst,rx_burst_hard , rx_burst_soft ] = viterbi_detector(SYMBOLS,NEXT,PREVIOUS,START,STOPS,Y,Rhh,Y1)
%
% VITERBI_DETECTOR:
%           This matlab code does the actual detection of the
%           received sequence. As indicated by the name the algorithm
%           is the viterbi algorithm, which is a MLSE. At this time 
%           the approch is to use Ungerboecks modified algorithm, and
%           to return hard output only.
%
% SYNTAX:   [ rx_burst ] 
%           = 
%           viterbi_detector(SYMBOLS,NEXT,PREVIOUS,START,STOPS,Y,Rhh)
%
% INPUT:    SYMBOLS:  The table of symbols corresponding the the state-
%                     numbers. Format as made by make_symbols.m
%           NEXT:     A transition table containing the next legal
%                     states, as it is generated by the code make_next.m
%           PREVIOUS: The transition table describing the legal previous
%		      states as generated by make_previous.m
%           START:    The start state of the algorithm.
%           STOPS:    The legal stop states.
%           Y:        Complex baseband representation of the matched 
%                     filtered and down converted received signal, as it
%                     is returned by mf.m
%           Rhh:      The autocorrelation as estimated by mf.m
%
% OUTPUT:   rx_burst: The most likely sequence of symbols.
%
% SUB_FUNC: make_increment
%
% WARNINGS: None.
%
% TEST(S):  Tested with no noise, perfect syncronization, and channel 
%           estimation/filtering. (Refer to viterbi_ill.m)
%
% % KNOWLEDGE OF Lh AND M IS NEEDED FOR THE ALGORITHM TO OPERATE

[ M , Lh ] = size(SYMBOLS);

% THE NUMBER OF STEPS IN THE VITERBI
%
STEPS=length(Y);

% INITIALIZE TABLES (THIS YIELDS A SLIGHT SPEEDUP).
%
METRIC = zeros(M,STEPS);
SURVIVOR = zeros(M,STEPS);
LLR = zeros(1,STEPS);  %  20070926

% DETERMINE PRECALCULATABLE PART OF METRIC
%
INCREMENT=make_increment(SYMBOLS,NEXT,Rhh);

% THE FIRST THING TO DO IS TO ROLL INTO THE ALGORITHM BY SPREADING OUT 
% FROM 	THE START STATE TO ALL THE LEGAL STATES. 
%
PS=START;

% NOTE THAT THE START STATE IS REFERRED TO AS STATE TO TIME 0
% AND THAT IT HAS NO METRIC.
%
S=NEXT(START,1);
METRIC(S,1)=real(conj(SYMBOLS(S,1))*Y(1))-INCREMENT(PS,S);
SURVIVOR(S,1)=START;

S=NEXT(START,2);
METRIC(S,1)=real(conj(SYMBOLS(S,1))*Y(1))-INCREMENT(PS,S);
SURVIVOR(S,1)=START;

PREVIOUS_STATES=NEXT(START,:);

% MARK THE NEXT STATES AS REAL. N.B: COMPLEX INDICATES THE POLARITY
% OF THE NEXT STATE, E.G. STATE 2 IS REAL.
%
COMPLEX=0;

for N = 2:Lh,
  if COMPLEX,
    COMPLEX=0;
  else
    COMPLEX=1;
  end
  STATE_CNTR=0;
  for PS = PREVIOUS_STATES,
    STATE_CNTR=STATE_CNTR+1;
    S=NEXT(PS,1);
    METRIC(S,N)=METRIC(PS,N-1)+real(conj(SYMBOLS(S,1))*Y(N))-INCREMENT(PS,S);
    SURVIVOR(S,N)=PS;
    USED(STATE_CNTR)=S;
    STATE_CNTR=STATE_CNTR+1;
    S=NEXT(PS,2);
    METRIC(S,N)=METRIC(PS,N-1)+real(conj(SYMBOLS(S,1))*Y(N))-INCREMENT(PS,S);
    SURVIVOR(S,N)=PS;    
    USED(STATE_CNTR)=S;
  end
  PREVIOUS_STATES=USED;
end
% AT ANY RATE WE WILL HAVE PROCESSED Lh STATES AT THIS TIME
%
PROCESSED=Lh;

% WE WANT AN EQUAL NUMBER OF STATES TO BE REMAINING. THE NEXT LINES ENSURE
% THIS.
%

if ~COMPLEX,
  COMPLEX=1;
  PROCESSED=PROCESSED+1;
  N=PROCESSED;
  for S = 2:2:M,
    PS=PREVIOUS(S,1);
    M1=METRIC(PS,N-1)+real(conj(SYMBOLS(S,1))*Y(N)-INCREMENT(PS,S));
    PS=PREVIOUS(S,2);
    M2=METRIC(PS,N-1)+real(conj(SYMBOLS(S,1))*Y(N)-INCREMENT(PS,S)); 
    if M1 > M2,
      METRIC(S,N)=M1;
      SURVIVOR(S,N)=PREVIOUS(S,1);
    else
      METRIC(S,N)=M2;
      SURVIVOR(S,N)=PREVIOUS(S,2);
    end
  end
end

% NOW THAT WE HAVE MADE THE RUN-IN THE REST OF THE METRICS ARE
% CALCULATED IN THE STRAIGHT FORWARD MANNER. OBSERVE THAT ONLY
% THE RELEVANT STATES ARE CALCULATED, THAT IS REAL FOLLOWS COMPLEX
% AND VICE VERSA.
%
N=PROCESSED+1;
while N <= STEPS,
  for S = 1:2:M-1,
    PS=PREVIOUS(S,1);
    M1=METRIC(PS,N-1)+real(conj(SYMBOLS(S,1))*Y(N)-INCREMENT(PS,S));
    PS=PREVIOUS(S,2);
    M2=METRIC(PS,N-1)+real(conj(SYMBOLS(S,1))*Y(N)-INCREMENT(PS,S)); 
    if M1 > M2,
      METRIC(S,N)=M1;
      SURVIVOR(S,N)=PREVIOUS(S,1);     
    else
      METRIC(S,N)=M2;
      SURVIVOR(S,N)=PREVIOUS(S,2);
    end
  end
  N=N+1;
  for S = 2:2:M,
    PS=PREVIOUS(S,1);
    M1=METRIC(PS,N-1)+real(conj(SYMBOLS(S,1))*Y(N)-INCREMENT(PS,S));
    PS=PREVIOUS(S,2);
    M2=METRIC(PS,N-1)+real(conj(SYMBOLS(S,1))*Y(N)-INCREMENT(PS,S)); 
    if M1 > M2,
      METRIC(S,N)=M1;
      SURVIVOR(S,N)=PREVIOUS(S,1);     
    else
      METRIC(S,N)=M2;
      SURVIVOR(S,N)=PREVIOUS(S,2);
    end
  end
  N=N+1;
end  


% HAVING CALCULATED THE METRICS, THE MOST PROBABLE STATESEQUENCE IS
% INITIALIZED BY CHOOSING THE HIGHEST METRIC AMONG THE LEGAL STOP 
% STATES.
%
BEST_LEGAL=0;
for FINAL = STOPS,
  if METRIC(FINAL,STEPS) > BEST_LEGAL,
    S=FINAL;
    BEST_LEGAL=METRIC(FINAL,STEPS);
  end
end

% UNCOMMENT FOR TEST OF METRIC
%
% METRIC
% BEST_LEGAL
% S
% pause

% HAVING FOUND THE FINAL STATE, THE MSK SYMBOL SEQUENCE IS ESTABLISHED
%
IEST(STEPS)=SYMBOLS(S,1);
N=STEPS-1;
while N > 0,
  S=SURVIVOR(S,N+1);
  IEST(N)=SYMBOLS(S,1);
  N=N-1;
end