Code for JT4 deep search now working well. (Needs some fine tuning, though.)

Message averaging has been tested, works well also.
Next: need to integrate these features and back-port into WSJT9.


git-svn-id: svn+ssh://svn.code.sf.net/p/wsjt/wsjt/branches/wsjtx@2969 ab8295b8-cf94-4d9e-aec4-7959e3be5d79

lib/chkmsg.f90

@@ -0,0 +1,31 @@
+subroutine chkmsg(message,cok,nspecial,flip)
+
+  character message*22,cok*3
+
+  nspecial=0
+  flip=1.0
+  cok="   "
+
+  do i=22,1,-1
+     if(message(i:i).ne.' ') go to 10
+  enddo
+  i=22
+
+10 if(i.ge.11) then
+     if((message(i-3:i).eq.' OOO') .or. (message(20:22).eq.' OO')) then
+        cok='OOO'
+        flip=-1.0
+        if(message(20:22).eq.' OO') then
+           message=message(1:19)
+        else
+           message=message(1:i-4)
+        endif
+     endif
+  endif
+
+  if(message(1:2).eq.'RO')  nspecial=2
+  if(message(1:3).eq.'RRR') nspecial=3
+  if(message(1:2).eq.'73')  nspecial=4
+
+  return
+end subroutine chkmsg

lib/decode24.f90

@@ -1,17 +1,20 @@
 subroutine decode24(dat,npts,dtx,dfx,flip,mode,mode4,decoded,ncount,   &
-     nlim,deepmsg,qual,submode)
+     deepmsg,qual,submode)
 
 ! Decodes JT65 data, assuming that DT and DF have already been determined.
 
   parameter (MAXAVE=120)
   real dat(npts)                        !Raw data
   character decoded*22,deepmsg*22
+  character*12 mycall,hiscall
+  character*6 hisgrid
   character*72 c72
   character submode*1
   real*8 dt,df,phi,f0,dphi,twopi,phi1,dphi1
   complex*16 cz,cz1,c0,c1
   integer*1 symbol(207)
   real*4 rsymbol(207,7)
+  real*4 sym(207)
   integer nsum(7)
   integer*1 data1(13)                   !Decoded data (8-bit bytes)
   integer   data4a(9)                   !Decoded data (8-bit bytes)
@@ -46,6 +49,9 @@ subroutine decode24(dat,npts,dtx,dfx,flip,mode,mode4,decoded,ncount,   &
   if(istart.lt.0) istart=0
   nchips=0
   ich=0
+  qbest=0.
+  deepmsg='                      '
+  ichbest=-1
 
 ! Should amp be adjusted according to signal strength?
 
@@ -103,20 +109,23 @@ subroutine decode24(dat,npts,dtx,dfx,flip,mode,mode4,decoded,ncount,   &
      if(i4.gt.127) i4=i4-256
      if(j.ge.1) then
         symbol(j)=i4
-        rsymbol(j,ich)=rsymbol(j,ich) + rsym
+!        rsymbol(j,ich)=rsymbol(j,ich) + rsym
+        rsymbol(j,ich)=rsym
+        sym(j)=rsym
      endif
   enddo
   
 !###  The following does simple message averaging:
-  nsum(ich)=nsum(ich)+1
+!  nsum(ich)=nsum(ich)+1
-  do j=1,207
+!  do j=1,207
-     r=rsymbol(j,ich)/nsum(ich) + 128.
+!     sym(j)=rsymbol(j,ich)/nsum(ich)
-     if(r.gt.255.0) r=255.0
+!     r=sym(j) + 128.
-     if(r.lt.0.0) r=0.0
+!     if(r.gt.255.0) r=255.0
-     i4=nint(r)
+!     if(r.lt.0.0) r=0.0
-     if(i4.gt.127) i4=i4-256
+!     i4=nint(r)
-     symbol(j)=i4
+!     if(i4.gt.127) i4=i4-256
-  enddo
+!     symbol(j)=i4
+!  enddo
 !###
   
   nbits=72+31
@@ -129,6 +138,20 @@ subroutine decode24(dat,npts,dtx,dfx,flip,mode,mode4,decoded,ncount,   &
   call fano232(symbol(2),nbits,mettab,delta,limit,data1,ncycles,metric,ncount)
   nlim=ncycles/nbits
 
+!### Try deep search
+  qual=0.
+  neme=1
+  mycall='VK7MO'
+  hiscall='W5LUA'
+  hisgrid='EM13'
+  call deep24(sym(2),neme,flip,mycall,hiscall,hisgrid,decoded,qual)
+  if(qual.gt.qbest) then
+     qbest=qual
+     deepmsg=decoded
+     ichbest=ich
+  endif
+!###
+
   if(ncount.ge.0) go to 100
   if(mode.eq.7 .and. nchips.lt.mode4) go to 40
 
@@ -137,27 +160,26 @@ subroutine decode24(dat,npts,dtx,dfx,flip,mode,mode4,decoded,ncount,   &
      if(i4.lt.0) i4=i4+256
      data4a(i)=i4
   enddo
-!  call cs_lock('decode24')
   write(c72,1100) (data4a(i),i=1,9)
 1100 format(9b8.8)
   read(c72,1102) data4
 1102 format(12b6)
-!  call cs_unlock
 
   decoded='                      '
   submode=' '
   if(ncount.ge.0) then
      call unpackmsg(data4,decoded)
      submode=char(ichar('A')+ich-1)
+  else
+     decoded=deepmsg
+     submode=char(ichar('A')+ichbest-1)
+     qual=qbest
   endif
   if(decoded(1:6).eq.'000AAA') then
      decoded='***WRONG MODE?***'
      ncount=-1
   endif
 
-  qual=0.
-  deepmsg='                      '
-
 ! Save symbol spectra for possible decoding of average.
 
   return

lib/deep24.f90

@@ -1,9 +1,9 @@
-subroutine deep24(s3,neme,flip,mycall,hiscall,hisgrid,decoded,qual)
+subroutine deep24(sym,neme,flip,mycall,hiscall,hisgrid,decoded,qual)
 
 ! Have barely begun converting this from JT65 to JT4
 
   parameter (MAXCALLS=7000,MAXRPT=63)
-  real s3(64,63)
+  real*4 sym(206)
   character callsign*12,grid*4,message*22,hisgrid*6,c*1,ceme*3
   character*12 mycall,hiscall
   character mycall0*12,hiscall0*12,hisgrid0*6
@@ -12,10 +12,10 @@ subroutine deep24(s3,neme,flip,mycall,hiscall,hisgrid,decoded,qual)
   character*15 callgrid(MAXCALLS)
   character*180 line
   character*4 rpt(MAXRPT)
-  integer ncode(63,2*MAXCALLS + 2 + MAXRPT)
+  integer ncode(206)
+  real*4   code(206,2*MAXCALLS + 2 + MAXRPT)
   real pp(2*MAXCALLS + 2 + MAXRPT)
-  common/mrscom/ mrs(63),mrs2(63)
+!  common/c3com/ mcall3a
-  common/c3com/ mcall3a
 
   data neme0/-99/
   data rpt/'-01','-02','-03','-04','-05',          &
@@ -78,7 +78,7 @@ subroutine deep24(s3,neme,flip,mycall,hiscall,hisgrid,decoded,qual)
      j3=index(mycall,'/')                 ! j3>0 means compound mycall
      j4=index(callsign,'/')               ! j4>0 means compound hiscall
      callgrid(icall)=callsign(1:j2)
-     
+
      mz=1
 ! Allow MyCall + HisCall + rpt (?)
      if(n.eq.1 .and. j3.lt.1 .and. j4.lt.1 .and.                       &
@@ -89,15 +89,16 @@ subroutine deep24(s3,neme,flip,mycall,hiscall,hisgrid,decoded,qual)
         message=mycall(1:j1)//' '//callgrid(icall)
         k=k+1
         testmsg(k)=message
-!        call encode65(message,ncode(1,k))
+        call encode4(message,ncode)
-        
+        code(1:206,k)=2*ncode(1:206)-1
         if(n.ge.2) then
 ! Insert CQ message
            if(j4.lt.1) callgrid(icall)=callsign(1:j2)//' '//grid
            message='CQ '//callgrid(icall)
            k=k+1
            testmsg(k)=message
-!           call encode65(message,ncode(1,k))
+           call encode4(message,ncode)
+        code(1:206,k)=2*ncode(1:206)-1
         endif
      enddo
 10   continue
@@ -110,10 +111,13 @@ subroutine deep24(s3,neme,flip,mycall,hiscall,hisgrid,decoded,qual)
 30 mycall0=mycall
   hiscall0=hiscall
   hisgrid0=hisgrid
-  ref0=0.
+
-  do j=1,63
+  sq=0.
-     ref0=ref0 + s3(mrs(j),j)
+  do j=1,206
+     sq=sq + sym(j)**2
   enddo
+  rms=sqrt(sq/206.0)
+  sym=sym/rms
 
   p1=-1.e30
   p2=-1.e30
@@ -121,14 +125,11 @@ subroutine deep24(s3,neme,flip,mycall,hiscall,hisgrid,decoded,qual)
      pp(k)=0.
 ! Test all messages if flip=+1; skip the CQ messages if flip=-1.
      if(flip.gt.0.0 .or. testmsg(k)(1:3).ne.'CQ ') then
-        sum=0.
+        p=0.
-        ref=ref0
+        do j=1,206
-        do j=1,63
+           i=code(j,k)+1
-           i=ncode(j,k)+1
+           p=p + code(j,k)*sym(j)
-           sum=sum + s3(i,j)
-           if(i.eq.mrs(j)) ref=ref - s3(i,j) + s3(mrs2(j),j)
         enddo
-        p=sum/ref
         pp(k)=p
         if(p.gt.p1) then
            p1=p
@@ -142,15 +143,20 @@ subroutine deep24(s3,neme,flip,mycall,hiscall,hisgrid,decoded,qual)
   enddo
 
 ! ### DO NOT REMOVE ### 
-  rewind 77
+!  rewind 77
-  write(77,*) p1,p2
+!  write(77,*) p1,p2
 ! ### Works OK without it (in both Windows and Linux) if compiled 
 ! ### without optimization.  However, in Windows this is a colossal 
 ! ### pain because of the way F2PY wants to run the compile step.
 
 
+  bias=1.1*p2
+!  if(mode65.eq.1) bias=max(1.12*p2,0.335)
+!  if(mode65.eq.2) bias=max(1.08*p2,0.405)
+!  if(mode65.ge.4) bias=max(1.04*p2,0.505)
+
   if(p2.eq.p1 .and. p1.ne.-1.e30) stop 'Error in deep24'
-  qual=100.0*(p1-bias)
+  qual=10.0*(p1-bias)
 
   decoded='                      '
   c=' '
@@ -169,5 +175,8 @@ subroutine deep24(s3,neme,flip,mycall,hiscall,hisgrid,decoded,qual)
           decoded(i:i)=char(ichar(decoded(i:i))-32)
   enddo
 
+!  write(*,3010) p1,p2,p1-p2,p1/p2,qual,decoded
+!3010 format('DS:',5f9.1,2x,a22)
+
   return
 end subroutine deep24

lib/encode4.f90

@@ -0,0 +1,20 @@
+subroutine encode4(message,ncode)
+
+  parameter (MAXCALLS=7000,MAXRPT=63)
+  integer ncode(206)
+  character*22 message          !Message to be generated
+  character*3 cok               !'   ' or 'OOO'
+  integer dgen(13)
+  integer*1 data0(13),symbol(216)
+  logical text
+
+  call chkmsg(message,cok,nspecial,flip)
+  call packmsg(message,dgen,text)  !Pack 72-bit message into 12 six-bit symbols
+  call entail(dgen,data0)
+  call encode232(data0,206,symbol)       !Convolutional encoding
+  call interleave24(symbol,1)            !Apply JT4 interleaving
+  do i=1,206
+     ncode(i)=symbol(i)
+  enddo
+
+end subroutine encode4

lib/gen24.f90

@@ -0,0 +1,86 @@
+subroutine gen24(message,mode4,samfac,ntxdf,iwave,nwave,sendingsh,msgsent,nmsg)
+
+! Encode a JT4 message into a wavefile.
+
+  parameter (NMAX=60*11025)     !Max length of wave file
+  character*22 message          !Message to be generated
+  character*22 msgsent          !Message as it will be received
+  character*3 cok               !'   ' or 'OOO'
+  real*8 t,dt,phi,f,f0,dfgen,dphi,pi,twopi,samfac,tsymbol
+  integer*2 iwave(NMAX)         !Generated wave file
+  integer sendingsh
+  integer dgen(13)
+  integer*1 data0(13),symbol(216)
+  logical first
+  include 'prcom2.f'
+  data first/.true./
+  save
+
+  nsym=207                               !Symbols per transmission
+  if(first) then
+     do i=1,nsym
+        pr2(i)=2*npr2(i)-1
+     enddo
+     pi=4.d0*atan(1.d0)
+     twopi=2.d0*pi
+     first=.false.
+  endif
+
+  call chkmsg(message,cok,nspecial,flip)
+  call packmsg(message,dgen)  !Pack 72-bit message into 12 six-bit symbols
+  call entail(dgen,data0)
+  call unpackmsg(dgen,msgsent)
+
+  nbytes=(72+31+7)/8
+  call encode(data0,nbytes,symbol(2))    !Convolutional encoding
+  symbol(1)=0                            !Reference phase
+  sendingsh=0
+  if(iand(dgen(10),8).ne.0) sendingsh=-1 !Plain text flag
+  call interleave24(symbol(2),1)         !Apply JT4 interleaving
+
+! Set up necessary constants
+  tsymbol=2520.d0/11025.d0
+  dt=1.d0/(samfac*11025.d0)
+  f0=118*11025.d0/1024 + ntxdf
+  dfgen=11025.d0/2520                     !4.375 Hz
+  t=0.d0
+  phi=0.d0
+  j0=0
+  ndata=(nsym*11025.d0*samfac*tsymbol)/2
+  ndata=2*ndata
+  do i=1,ndata
+     t=t+dt
+     j=int(t/tsymbol) + 1   !Symbol number, 1-207
+     if(j.ne.j0) then
+        f=f0 + (npr2(j)+2*symbol(j)-1.5) * dfgen * mode4
+        if(flip.lt.0.0) f=f0+((1-npr2(j))+2*symbol(j)-1.5)*dfgen*mode4
+        dphi=twopi*dt*f
+        j0=j
+     endif
+     phi=phi+dphi
+     iwave(i)=32767.0*sin(phi)
+  enddo
+
+  do j=1,5512                !Put another 0.5 sec of silence at end
+     i=i+1
+     iwave(i)=0
+  enddo
+  nwave=i
+
+  if(flip.lt.0.0) then
+     do i=22,1,-1
+        if(msgsent(i:i).ne.' ') goto 10
+     enddo
+10   msgsent=msgsent(1:i)//' OOO'
+  endif
+  do i=22,1,-1
+     if(msgsent(i:i).ne.' ') goto 20
+  enddo
+20 nmsg=i
+
+!      write(*,3002) (symbol(i),i=1,207)
+! 3002 format(70i1)
+
+  return
+end subroutine gen24
+

lib/wsjt24.f90

@@ -92,7 +92,7 @@ subroutine wsjt24(dat,npts,cfile6,NClearAve,MinSigdB,                  &
   endif
 
   call decode24(dat,npts,dtx,dfx,flip,mode,mode4,decoded,           &
-       ncount,nlim,deepmsg,qual,submode)
+       ncount,deepmsg,qual,submode)
 
 200 kvqual=0
   if(ncount.ge.0) kvqual=1
@@ -112,8 +112,8 @@ subroutine wsjt24(dat,npts,cfile6,NClearAve,MinSigdB,                  &
 
 !  call cs_lock('wsjt24')
   write(line,1010) cfile6,nsync,nsnr,dtx-1.0,jdf,nint(width),         &
-       csync,special,decoded(1:19),cooo,kvqual,nqual,submode,nlim
+       csync,special,decoded(1:19),cooo,kvqual,nqual,submode
-1010 format(a6,i3,i5,f5.1,i5,i3,1x,a1,1x,a5,a19,1x,a3,i4,i4,1x,a1,i8)
+1010 format(a6,i3,i5,f5.1,i5,i3,1x,a1,1x,a5,a19,1x,a3,i3,i5,1x,a1)
 
 ! Blank all end-of-line stuff if no decode
   if(line(31:40).eq.'          ') line=line(:30)

lib/wsjt24d.f90

@@ -5,6 +5,7 @@ program wsjt24d
   character*12 arg
   real ccfblue(-5:540)        !X-cor function in JT65 mode (blue line)
   real ccfred(450)            !Average spectrum of the whole file
+  integer dftolerance
 
   nargs=iargc()
   if(nargs.ne.2) then
@@ -16,6 +17,7 @@ program wsjt24d
   call getarg(2,arg)
   read(arg,*) ifile2
 
+  open(23,file='CALL3.TXT',status='old')
   open(50,file='vk7mo.dat',form='unformatted',status='old')
 
   do ifile=1,ifile2
@@ -27,6 +29,10 @@ program wsjt24d
 !     write(*,3000) ifile,cfile6,jz,mode,mode4,idf
 !3000 format(i3,2x,a6,i10,3i5)
 
+     dftolerance=100
+     nfreeze=1
+     neme=0
+
 !    call wsjt24(dat(4097),jz-4096,cfile6,NClearAve,MinSigdB,DFTolerance,  &
     call wsjt24(dat,jz,cfile6,NClearAve,MinSigdB,DFTolerance,             &
          NFreeze,mode,mode4,Nseg,MouseDF2,NAgain,idf,lumsg,lcum,nspecial, &

mainwindow.cpp

@@ -1,4 +1,4 @@
-//-------------------------------------------------------------- MainWindow
+//------------------------------------------------------------- MainWindow
 #include "mainwindow.h"
 #include "ui_mainwindow.h"
 #include "devsetup.h"