mirror of
https://github.com/saitohirga/WSJT-X.git
synced 2024-11-16 00:51:56 -05:00
f398495cfc
The decoded message processing relies on fixed column positions so the extra character used in message averaging to indicate sync flips must also be allowed for in other modes. This is done by inserting a space character. git-svn-id: svn+ssh://svn.code.sf.net/p/wsjt/wsjt/branches/wsjtx@6681 ab8295b8-cf94-4d9e-aec4-7959e3be5d79
120 lines
3.9 KiB
Fortran
120 lines
3.9 KiB
Fortran
subroutine jtmsk_decode(id2,narg,line)
|
|
|
|
! Decoder for JTMSK mode
|
|
|
|
parameter (NMAX=30*12000)
|
|
parameter (NFFTMAX=512*1024)
|
|
parameter (NSPM=1404) !Samples per JTMSK long message
|
|
integer*2 id2(0:NMAX) !Raw i*2 data, up to T/R = 30 s
|
|
integer hist(0:32868)
|
|
real d(0:NMAX) !Raw r*4 data
|
|
real ty(NMAX/512) !Ping times
|
|
real yellow(NMAX/512)
|
|
complex c(NFFTMAX) !Complex (analytic) data
|
|
complex cdat(24000) !Short segments, up to 2 s
|
|
complex cdat2(24000)
|
|
integer narg(0:14) !Arguments passed from calling pgm
|
|
character*22 msg,msg0 !Decoded message
|
|
character*80 line(100) !Decodes passed back to caller
|
|
equivalence (hist,d)
|
|
|
|
! Parameters from GUI are in narg():
|
|
nutc=narg(0) !UTC
|
|
npts=min(narg(1),NMAX) !Number of samples in id2 (12000 Hz)
|
|
newdat=narg(3) !1==> new data, compute symbol spectra
|
|
minsync=narg(4) !Lower sync limit
|
|
npick=narg(5)
|
|
t0=0.001*narg(6)
|
|
t1=0.001*narg(7)
|
|
maxlines=narg(8) !Max # of decodes to return to caller
|
|
nmode=narg(9)
|
|
nrxfreq=narg(10) !Target Rx audio frequency (Hz)
|
|
ntol=narg(11) !Search range, +/- ntol (Hz)
|
|
nhashcalls=narg(12)
|
|
naggressive=narg(14)
|
|
nsnr0=-99
|
|
nline=0
|
|
line(1:100)(1:1)=char(0)
|
|
msg0=' '
|
|
msg=msg0
|
|
|
|
hist=0
|
|
do i=0,npts-1
|
|
n=abs(id2(i))
|
|
hist(n)=hist(n)+1
|
|
enddo
|
|
ns=0
|
|
do n=0,32768
|
|
ns=ns+hist(n)
|
|
if(ns.gt.npts/2) exit
|
|
enddo
|
|
fac=1.0/(1.5*n)
|
|
d(0:npts-1)=fac*id2(0:npts-1)
|
|
! rms=sqrt(dot_product(d(0:npts-1),d(0:npts-1))/npts)
|
|
!### Would it be better to set median rms to 1.0 ?
|
|
! d(0:npts-1)=d(0:npts-1)/rms !Normalize so that rms=1.0
|
|
call mskdt(d,npts,ty,yellow,nyel)
|
|
nyel=min(nyel,5)
|
|
|
|
n=log(float(npts))/log(2.0) + 1.0
|
|
nfft=min(2**n,1024*1024)
|
|
call analytic(d,npts,nfft,c) !Convert to analytic signal and filter
|
|
|
|
nbefore=NSPM
|
|
nafter=4*NSPM
|
|
! Process ping list (sorted by S/N) from top down.
|
|
do n=1,nyel
|
|
ia=ty(n)*12000.0 - nbefore
|
|
if(ia.lt.1) ia=1
|
|
ib=ia + nafter
|
|
if(ib.gt.NFFTMAX) ib=NFFTMAX
|
|
iz=ib-ia+1
|
|
cdat2(1:iz)=c(ia:ib) !Select nlen complex samples
|
|
ja=ia/NSPM + 1
|
|
jb=ib/NSPM
|
|
t0=ia/12000.0
|
|
|
|
do itry=1,21
|
|
idf1=(itry/2) * 50
|
|
if(mod(itry,2).eq.1) idf1=-idf1
|
|
if(abs(idf1).gt.ntol) exit
|
|
fpk=idf1 + nrxfreq
|
|
call tweak1(cdat2,iz,1500.0-fpk,cdat)
|
|
call syncmsk(cdat,iz,jpk,ipk,idf,rmax,snr,metric,msg)
|
|
if(metric.eq.-9999) cycle !No output if no significant sync
|
|
if(msg(1:1).eq.' ') call jtmsk_short(cdat,iz,narg,tbest,idfpk,msg)
|
|
if(msg(1:1).eq.'<' .and. naggressive.eq.0 .and. &
|
|
narg(13)/8.ne.narg(12)) msg=' '
|
|
if(msg(1:1).ne.' ') then
|
|
if(msg.ne.msg0) then
|
|
nline=nline+1
|
|
nsnr0=-99
|
|
endif
|
|
freq=fpk+idf
|
|
t0=(ia+jpk)/12000.0
|
|
y=10.0**(0.1*(yellow(n)-1.5))
|
|
nsnr=max(-5,nint(db(y)))
|
|
if(nsnr.gt.nsnr0 .and. nline.gt.0) then
|
|
call rectify_msk(cdat2(jpk:jpk+NSPM-1),msg,narg(13),freq2)
|
|
freq=freq2
|
|
if(msg(1:1).eq.'<') freq=freq2+idfpk
|
|
!### Check freq values !!!
|
|
write(line(nline),1020) nutc,nsnr,t0,nint(freq),msg
|
|
1020 format(i6.6,i4,f5.1,i5,' & ',a22)
|
|
nsnr0=nsnr
|
|
go to 900
|
|
endif
|
|
msg0=msg
|
|
if(nline.ge.maxlines) go to 900
|
|
endif
|
|
enddo
|
|
! print*,'c',nutc,n,nint(yellow(n)-4.0),freq,freq2
|
|
enddo
|
|
|
|
900 continue
|
|
! print*,'d',nutc,n,nint(yellow(n)-4.0),freq,freq2
|
|
if(line(1)(1:6).eq.' ') line(1)(1:1)=char(0)
|
|
|
|
return
|
|
end subroutine jtmsk_decode
|