| Printable Version of Topic
Click here to view this topic in its original format |
| Unofficial VirtualDub Support Forums > VirtualDub Filters and Filter Development > Any Way To Filter Out Echoes |
| Posted by: Arctonyx Mar 14 2003, 02:18 AM |
| One of my TV channels has atrocious ghosting. Many positive and negative reflection artefacts. This in spite of having an expensive antenna on my roof. The signal is bouncing off some hills and a huge grandstand. Is there some way to clean up the picture by filtering some of these out? I imagine that a one-dimensional implementation of a deconvolution algorythm, as discussed in a recent thread, might do the trick, but is there any currently existing filter or sequence of filters? Thanks |
| Posted by: Arctonyx Mar 28 2003, 10:07 PM |
| Would a comb filter do this? What is the effect of a comb filter anyway? From my understanding it can remove a specific frequency and it's harmonics from a signal. In audio it would be used for removing mains hum. But what is it's use in video processing? |
| Posted by: DD_DD Mar 31 2003, 12:58 PM |
| Hello, I don't know whether "comb" filter can perform what you want, but what about "exorcist" filter by Chris Wojdon. Short description of this filter: Exorcist - Developed to remove common aerial signal problem - shifted copies of main video. When you see two headed news speaker or 44 football players at one time running after two balls, you are in 99% victim of bad signal |
| Posted by: phaeron Apr 1 2003, 03:21 AM |
| Comb filters are named because their frequency response resembles a comb -- that is, notches or peaks at regular intervals. They do indeed allow you to isolate or remove a frequency and its harmonics. However, comb filters are usually much less exciting in implementation than they sound. For instance, comb filters are used to filter out the chrominance subcarrier (3.58MHz) when decoding composite video, as it would normally show up as a crawling checkerboard overlay. The line-rate comb filter for this is either [.5 .5] (average adjacent lines) or [.25 .5 .25] (3-tap triangle filter), while a frame-rate comb filter does the same on frames instead of lines. These filters aren't spectacular by themselves and you can get better results with adaptive switching between the two. The problem here is that the comb filter better resembles the filter kernel that caused the echo, not the one you need to remove it. Remember, you only have the echoed signal to cancel with, not the original signal. So when you try to cancel the echo by adding, you'll end up with a weaker secondary echo twice as far. As you may have guessed, the kernel you need to cancel this pattern is infinite... but your signal isn't. Some of it is lost in the horizontal blank and some by signal level clipping (overwrite or overblack). So the practical implementation of such an echo-cancellation filter is a bit more involved than the theory. |
| Posted by: Arctonyx Apr 2 2003, 01:09 PM |
| Thanks for the reference to Exorcist, DD_DD. And thanks for the explanation of the effect of comb filtering, phaeron. If the echoes were a strictly linear phenomenon then they could be de-convoluted out. But, as you point out, that isn't the case. Some nice food for thought there. |