WSJT-X/lib/msk144signalquality.f90

218 lines
6.1 KiB
Fortran
Raw Normal View History

First attempt at a UI phase compensation tool for MSK144 This builds on the static phase compensation in the MSK144 decoder and the phase analysis and polynomial fitting capabilities also in teh MSK144 decoder, by allowing captured data to be selected for phase equalization from the WSJT-X UI. Reads captured phase compensation curve estimate files containing fitted polynomial coefficients and measured phase data from MSK144 receptions. Intent is to select a compensation curve that is from a known transmitter like an SDR which have good phase linearity. Phase plots and compensation polynomials may be viewed and compared with the current compensation polynomial. A suitable polynomial can be applied to be use in all further decoding of MSK144 signals. Plots of the currently selected polynomial and its modified higher order terms polynomial which is actually used in equalization (this plot may be dropped - it is just for kicks at the moment). When a captured phase analysis file is loaded plots of the measured phase and the proposed best fit polynomial are shown. Basic maintenance is also included allowing clearing and loading captured plots and an option to revert to a flat no equalization curve. More to come on this as amplitude equalization is also possible, this will probably be similar, maybe even plotted on the same graph with dual axes for phase and amplitude. Amplitude correction from a measured reference spectrum could be viewed and selected for equalization for all modes. TBC... This change also introduces the QCustomPlot 3rd party widget. Currently this is statically linked from a qcp library built by the WSJT-X CMake script. This will probably be migrated to a shared object (DLL) build as a CMake external project, once some CMake script re-factoring has been completed, which is more in line with the QCustomPlot author's intentions. This will allow efficient reuse in other tools shipped with WSJT-X. git-svn-id: svn+ssh://svn.code.sf.net/p/wsjt/wsjt/branches/wsjtx@7570 ab8295b8-cf94-4d9e-aec4-7959e3be5d79
2017-02-20 21:13:13 -05:00
subroutine msk144signalquality(cframe,snr,freq,t0,softbits,msg,dxcall, &
btrain,datadir,nbiterrors,eyeopening,pcoeffs)
character*22 msg,msgsent
character*12 dxcall
character*12 training_dxcall
character*12 trained_dxcall
character*6 mygrid
character*512 pcoeff_filename
character*8 date
character*10 time
character*5 zone
character*512 datadir
complex cframe(864)
complex cross(864)
complex cross_avg(864)
complex canalytic(1024)
complex cmodel(1024)
integer i4tone(144)
integer hardbits(144)
integer msgbits(144)
integer values(8)
logical*1 bcontest
logical*1 btrain
logical*1 first
logical*1 currently_training
logical*1 msg_has_dxcall
logical*1 is_training_frame
real softbits(144)
real waveform(0:863)
real d(1024)
real phase(864)
real twopi,freq,phi,dphi0,dphi1,dphi
real*8 x(145),y(145),pp(145),sigmay(145),a(5),chisqr
First attempt at a UI phase compensation tool for MSK144 This builds on the static phase compensation in the MSK144 decoder and the phase analysis and polynomial fitting capabilities also in teh MSK144 decoder, by allowing captured data to be selected for phase equalization from the WSJT-X UI. Reads captured phase compensation curve estimate files containing fitted polynomial coefficients and measured phase data from MSK144 receptions. Intent is to select a compensation curve that is from a known transmitter like an SDR which have good phase linearity. Phase plots and compensation polynomials may be viewed and compared with the current compensation polynomial. A suitable polynomial can be applied to be use in all further decoding of MSK144 signals. Plots of the currently selected polynomial and its modified higher order terms polynomial which is actually used in equalization (this plot may be dropped - it is just for kicks at the moment). When a captured phase analysis file is loaded plots of the measured phase and the proposed best fit polynomial are shown. Basic maintenance is also included allowing clearing and loading captured plots and an option to revert to a flat no equalization curve. More to come on this as amplitude equalization is also possible, this will probably be similar, maybe even plotted on the same graph with dual axes for phase and amplitude. Amplitude correction from a measured reference spectrum could be viewed and selected for equalization for all modes. TBC... This change also introduces the QCustomPlot 3rd party widget. Currently this is statically linked from a qcp library built by the WSJT-X CMake script. This will probably be migrated to a shared object (DLL) build as a CMake external project, once some CMake script re-factoring has been completed, which is more in line with the QCustomPlot author's intentions. This will allow efficient reuse in other tools shipped with WSJT-X. git-svn-id: svn+ssh://svn.code.sf.net/p/wsjt/wsjt/branches/wsjtx@7570 ab8295b8-cf94-4d9e-aec4-7959e3be5d79
2017-02-20 21:13:13 -05:00
real pcoeffs(5)
data first/.true./
save cross_avg,wt_avg,first,currently_training, &
navg,tlast,training_dxcall,trained_dxcall
if (first) then
navg=0
cross=cmplx(0.0,0.0)
cross_avg=cmplx(0.0,0.0)
wt_avg=0.0
tlast=0.0
trained_dxcall(1:12)=' '
training_dxcall(1:12)=' '
currently_training=.false.
first=.false.
endif
if( (currently_training .and. (dxcall .ne. training_dxcall)) .or. &
(navg .gt. 10 )) then !reset and retrain
navg=0
cross=cmplx(0.0,0.0)
cross_avg=cmplx(0.0,0.0)
wt_avg=0.0
tlast=0.0
trained_dxcall(1:12)=' '
currently_training=.false.
training_dxcall(1:12)=' '
trained_dxcall(1:12)=' '
write(*,*) 'reset to untrained state '
endif
indx_dxcall=index(msg,trim(dxcall))
msg_has_dxcall = indx_dxcall .ge. 4
if( btrain .and. msg_has_dxcall .and. (.not. currently_training) ) then
currently_training=.true.
training_dxcall=trim(dxcall)
trained_dxcall(1:12)=' '
write(*,*) 'start training on call ',training_dxcall
endif
if( msg_has_dxcall .and. currently_training ) then
trained_dxcall(1:12)=' '
training_dxcall=dxcall
endif
! use decoded message to figure out how many bit errors in the frame
do i=1, 144
hardbits(i)=0
if(softbits(i) .gt. 0 ) hardbits(i)=1
enddo
! generate tones from decoded message
mygrid="EN50"
ichk=0
bcontest=.false.
call genmsk144(msg,mygrid,ichk,bcontest,msgsent,i4tone,itype)
! reconstruct message bits from tones
msgbits(1)=0
do i=1,143
if( i4tone(i) .eq. 0 ) then
if( mod(i,2) .eq. 1 ) then
msgbits(i+1)=msgbits(i)
else
msgbits(i+1)=mod(msgbits(i)+1,2)
endif
else
if( mod(i,2) .eq. 1 ) then
msgbits(i+1)=mod(msgbits(i)+1,2)
else
msgbits(i+1)=msgbits(i)
endif
endif
enddo
nbiterrors=0
do i=1,144
if( hardbits(i) .ne. msgbits(i) ) nbiterrors=nbiterrors+1
enddo
nplus=0
nminus=0
eyetop=1
eyebot=-1
do i=1,144
if( msgbits(i) .eq. 1 ) then
if( softbits(i) .lt. eyetop ) eyetop=softbits(i)
else
if( softbits(i) .gt. eyebot ) eyebot=softbits(i)
endif
enddo
eyeopening=eyetop-eyebot
is_training_frame = &
(snr.gt.5.0 .and.(nbiterrors.lt.7)) .and. &
(abs(t0-tlast) .gt. 0.072) .and. &
msg_has_dxcall
if( currently_training .and. is_training_frame ) then
twopi=8.0*atan(1.0)
nsym=144
if( i4tone(41) .lt. 0 ) nsym=40
dphi0=twopi*(freq-500)/12000.0
dphi1=twopi*(freq+500)/12000.0
phi=-twopi/8
indx=0
do i=1,nsym
if( i4tone(i) .eq. 0 ) then
dphi=dphi0
else
dphi=dphi1
endif
do j=1,6
waveform(indx)=cos(phi);
indx=indx+1
phi=mod(phi+dphi,twopi)
enddo
enddo
! convert the passband waveform to complex baseband
npts=864
nfft=1024
d=0
d(1:864)=waveform(0:863)
call analytic(d,npts,nfft,canalytic,pcoeffs,.false.,.false.) ! don't equalize the model
call tweak1(canalytic,nfft,-freq,cmodel)
call four2a(cframe(1:864),864,1,-1,1)
call four2a(cmodel(1:864),864,1,-1,1)
! Cross spectra from different messages can be averaged
! as long as all messages originate from dxcall.
cross=cmodel(1:864)*conjg(cframe)/1000.0
cross=cshift(cross,864/2)
cross_avg=cross_avg+10**(snr/20.0)*cross
wt_avg=wt_avg+10**(snr/20.0)
navg=navg+1
tlast=t0
phase=atan2(imag(cross_avg),real(cross_avg))
df=12000.0/864.0
nm=145
do i=1,145
x(i)=(i-73)*df/1000.0
enddo
y=phase((864/2-nm/2):(864/2+nm/2))
sigmay=wt_avg/abs(cross_avg((864/2-nm/2):(864/2+nm/2)))
mode=1
npts=145
nterms=5
call polyfit(x,y,sigmay,npts,nterms,mode,a,chisqr)
pp=a(1)+x*(a(2)+x*(a(3)+x*(a(4)+x*a(5))))
rmsdiff=sum( (pp-phase((864/2-nm/2):(864/2+nm/2)))**2 )/145.0
write(*,*) 'training ',navg,sqrt(chisqr),rmsdiff
if( (sqrt(chisqr).lt.1.8) .and. (rmsdiff.lt.0.5) .and. (navg.ge.5) ) then
trained_dxcall=dxcall
training_dxcall(1:12)=' '
currently_training=.false.
btrain=.false.
call date_and_time(date,time,zone,values)
write(pcoeff_filename,'(i2.2,i2.2,i2.2,"_",i2.2,i2.2,i2.2)') &
values(1)-2000,values(2),values(3),values(5),values(6),values(7)
pcoeff_filename=trim(trained_dxcall)//"_"//trim(pcoeff_filename)//".pcoeff"
l1=index(datadir,char(0))-1
datadir(l1+1:l1+1)="/"
pcoeff_filename=datadir(1:l1+1)//trim(pcoeff_filename)
write(*,*) 'trained - writing coefficients to: ',pcoeff_filename
open(17,file=pcoeff_filename,status='new')
write(17,'(i4,2f10.2,5f10.4)') navg,sqrt(chisqr),rmsdiff,a(1),a(2),a(3),a(4),a(5)
do i=1, 145
write(17,*) x(i),pp(i),y(i),sigmay(i)
enddo
do i=1,864
write(17,*) i,real(cframe(i)),imag(cframe(i)),real(cross_avg(i)),imag(cross_avg(i))
enddo
close(17)
endif
endif
return
end subroutine msk144signalquality