itewreed / secam-tbc-2-yuv

GNU radio based graph to convert secam based TBC files of the vhs-decode project to a YUV file

secam-tbc-2-yuv

GNU radio based graph to convert SECAM based TBC files of the vhs-decode project to a YUV file. Combination with original mkv file possible to benefit from dropout correction. The graph is based on Gnuradio 3.10

Quick usage guide

• Load both tbc files into the graph, set output file
• Run graph, adjust invert_crcb parameter (value 0 or 1) if necessary
• Use bash script to combine with dropout corrected mkv file create by gen_chroma_vid script

Graphs

Color values are set to properly decode color of TBC files who were decoded by using the MESECAM option in vhs-decode. Also MESECAM option should be the preferred one for decoding, as there are less color streaks in the decoded video.

CLI Usage

The Python has to be used that GNURadio also uses to execute the file. Should usually be in /usr/bin/python3. GNURadio has to be installed to make the Python script works.

/usr/bin/python3 -u  secam2yuv.py  --inputdir <directory where the tbc files are> --videofile <filename without .tbc> --outputdir <directory to save the bin file to> --invert_crcb <0 or 1>
# example
/usr/bin/python3 -u  secam2yuv.py  --inputdir /home/vhsuser/tbc/ --videofile tape477_01_visio_1991-06-09_16mhz --outputdir /home/vhsuser/Videos/ --invert_crcb 1

Usage

Load the graph in GnuRadio Companion. Set the inputdir, outputdir and videofile parameters. The output file name is the name of the input + info if invert was set aka videofile_i0.bin. Run the graph. There is a color video preview in the graph. It will slow down the conversion, but it's highly recommend to use it, as you can live check if the video flaps to pink at some point. If it flaps, quickly press the invert_crcb_switch button, then colors should be back to normal

In the next step use ffmpeg to convert that raw YUV file.

ffmpeg -f rawvideo -pixel_format yuv444p -color_range tv -color_primaries "bt470bg" -color_trc "gamma28" -colorspace "bt470bg" -video_size 1185x624 -r 25 -i YUV.bin -filter:v "crop=928:576:183:46, setdar=1856/1383" -c:v ffv1 -coder 1 -context 1 -g 25 -level 3 -slices 16 -slicecrc 1 -top 1 output.mkv

It also crops the video to the same dimensions and cutout as it would be from the gen_chroma_vid script. This way it is possible to merge the chroma and luma planes from the normal ld-chroma-decoder and the gnuradio graph output in order to benefit from dropout correction.

The video dimensions of the file is 928x576, with 25fps. Rendering out as ffv1 is recommended. The output should be a proper interlaced video, thus can be properly deinterlaced if needed.

Merge color with dropout corrected MKV file

AviSynth way

Use the AviSynth script. The additional module RawSourcePlus needs to be installed beforehand. Using AVSPmod a live preview is available. Put in the location of the videofiles, ld muxed MKV file into FFmpegSource2 and GNURadio created bin file into RawSourcePlus. The trimming needs to be commented in if startfield is still used and startfield is 2, otherwise it needs to be commented out. Save and open the script in your desired video application that understands AviSynth scripts.

CLI Way

1. Create the YUV file with the GNURadio graph
2. Render this as MKV file as explained in this docu
3. create the video file as usual with the gen_chroma_vid script from vhs / ld-decode project
4. Merge the files and planes using the following command

ffmpeg -y -i "videofromgnuradio" -ss 0.00 -i "videofromgenvidscript  " -filter_complex " [0:v]format=yuv444p[0v]; [1:v]format=yuv422p10le[1v]; [0v][1v]mergeplanes=0x100102:yuv422p10le[v]" -map 1:a -c:a copy -map '[v]' -an -c:v ffv1 -coder 1 -context 1 -g 25 -level 3 -slices 16 -slicecrc 1 -top 1 "video_merged.mkv"

The order is important, as the mapping is as follows in the output. Y of 2nd input -> Y, U of 1st input -> U, V of 1st input -> V

You can also omit one step and directly merge the YUV file from Gnuradio with the gen_chroma_vid generated file.

ffmpeg -y -f rawvideo -pixel_format yuv444p -color_range tv -color_primaries "bt470bg" -color_trc "gamma28" -colorspace "bt470bg" -video_size 1185x624 -r 25 -i "videofromgnuradio" -ss 0.00 -i "videofromgenvidscript" -filter_complex "[0:v]format=yuv444p, crop=928:576:183:46, setdar=1856/1383[chroma]; [1:v]format=yuv422p10le, setdar=1856/1383[luma]; [chroma][luma]mergeplanes=0x100102:yuv422p10le[v]" -map 1:a -c:a copy -map '[v]' -c:v ffv1 -coder 1 -context 1 -g 25 -level 3 -slices 16 -slicecrc 1 -top 1 "video_merged.mkv"

In both cases the -ss option is important to align both files. Set it from 0.00 to 0.04 if you still use startfield and your startfield is 2!
0.04 means 1/25th of a second, representing skipping one frame at the beginning from the gen_chroma_vid generated file.

Limitations

This is an experimental solution to decode SECAM video from the vhs / ld-decode project. A flaw where the right border is pink has been fixed with a workaround. Depending on the input file also the whole video window can get overpainted by this. The cause and a solution is yet unkown. The demodulation and filtering of color is very basic. There can be a lot of sparks in the color (secam fire), depending on how good or bad the SNR of the decoded input is. No dropout compensation is present, as this is something usually done by the ld-decode components. But it can be added for the luma (Y) component, see the merge color section.

Flaws present can be, to a certain degree, alleviated in post.

Troubleshooting

If the majority of the video is affected by pink overpainting there are two possible solutions to fix that.

1. Give the input files an offset

To shift the offset by one frame we have to take account, that one field in this graph is 1135 samples per line over 626 lines. So the formula is (1135*626)*startfield.
The startfield is a variable that should be toggled between 0 and 2. Theoretically it can be anything between zero and the last field of the video. So use this to find a start frame where both fields are not pink. (As said usually only 0 and 2 make sense)

2. Decode by using --start parameter

Try to decode with --start option in vhs-decode. This will start the decode from another frame and reverts this effect. You might have to try out several start frames for success. But this is less convenient than option 1.