Sunday, October 4, 2015

ATSC update

Hi again. While making ATSC recordings, I ran into a problem that I think you should be aware of. Even with 100% perfect reception, your final decodes may be corrupt.

The moral of this post: Use RAM drives to record your IQ samples! SSD's might also be a good idea.

How I found out:
As I described in the first post, it takes a lot of bandwidth to record the IQ samples necessary to capture a TV station. Surprisingly, it takes 32 MiB/sec for 8 MHz of bandwidth, even though you're doing 16-bit samples. It seems that it should be 8 Megasamples/sec * 2 bytes/sample = 16 Megabytes/second.

If you use a magnetic hard drive to record IQ samples and some of your TV recordings are corrupt and won't play, chances are good it's the hard drive not keeping up.

On a side note: you may have also noticed that 8 MHz looks cleaner in the waterfall than 7 MHz, although 7 MHz can work. 6 MHz is just too narrow, because of the IF filters that round off the edges.

Using RAM drives, I had been limited to ~1500-1800 MB available RAM on one computer, which was only enough for about a minute. The only way to get more space was to use a hard drive. But hard drive recordings never played very well. I began to wonder if the recordings somehow got bad after a certain amount of time.

I tested this by taking several very large IQ recordings (of different stations.) Some were recorded to an 8 TB external drive I purchased recently, others to a very fast internal magnetic SATA drive, and one more on a 5 or 6 GB RAM drive (on different computer.)

Of the three, the only one I could actually watch was the one recorded to the RAM drive.

This is proof enough that filesize won't mess up the decoder; apparently the hard drives were causing samples to be dropped.


  1. Hi,

    Congratulations for your effort on this topic. It is very interesting. Do you believe that is possible decode ATSC real time?

    Best Regards,
    Claudio F. Dias

  2. Yes, it is possible, but requires a very fast computer. Ron Economos said it is 0.5x real-time on his Intel Xeon E5-1607.

    I read that Johnathan Corgan was able to do it on a 24-core computer. You can read more at