The Call of the Open Sidewalk

Digital television (DTV) works a lot better than analog television for the case where the signal gets broadcast through the atmosphere (OTA). There are 4 major standards for broadcasting TV digitally. The profound political failure to settle on a single standard was as usual in these sorts of cases due mostly to pointless nationalism and greed. The system adopted in the Americas (where I live) is called ATSC. This system is (controversially) different from the others in that it uses amplitude modulation (AM) as the way to represent the bitstream. Often times different is also interesting...

In a previous post I claimed that analog TV was bad because it used AM modulation and was thus very susceptible to noise. The ATSC DTV system uses AM modulation but works very well and is quite noise resistant. There are two major reasons for this.

The most interesting of the two reasons is called forward error correction. Let's tell the story in a way appropriate to the medium under discussion:

INT. CHIEF'S OFFICE - NIGHT

The DETECTIVE bursts into the office.

DETECTIVE There is something wrong with the bits!

CHIEF There can't be.

DETECTIVE Have a look! You know the encoder. Is this something she would do?

CHIEF No. I mean that they *can't* be wrong. That data is heavily compressed. If bad data hits the decompressor there will be hell to pay. ... Who did this?

DETECTIVE Terrorists. Definitely terrorists.

CHIEF (perfectly credulous) Of course! What if anything can we do about it?

DETECTIVE (thoughtfully) Well bits only have two states. If we can figure out which bits are wrong we can just change their states and everything will be fine.

CHIEF We just don't have the time for all that! I'll call the mayor.

DETECTIVE No! Wait! Bit 356 has to be wrong.

CHIEF Well obviously. It is just as obvious that that is not the whole story... Give it up. This case is already over.

DETECTIVE (desperate) I've got it! Bits 356, 790 and 1393.

CHIEF Three? Now you are just clutching at straws. Look. I'm dialing the phone...

DETECTIVE (still desperate) Look at it!

The CHIEF looks.

CHIEF Well what do you know? You figured it out.

The CHIEF inexplicably pulls a shotgun out from under his desk. He stands up. He racks the slide.

CHIEF (forcefully) Looks like we got us some bits to flip!

Forward error correction (FEC) works by adding extra data to the data you want to transmit. That extra data is made to be dependent on the transmitted data in a deliberately complex way so that the receiver can figure out with a high degree of probability which bits are different from what was sent. I am not really anthropomorphizing here. A television with a ATSC tuner is doing something that would be called deductive reasoning if a person did it. Really difficult deductive reasoning. The TV does this constantly and could not do without this capability.

ATSC actually uses two types of FEC, one after the other. Trellis modulation (famous for making telephone modems practical) and Reed–Solomon error correction (famous for making Compact Disks practical). Around one third of all the bits transmitted are for the purpose of error correction. This is a high cost but the data an ATSC receiver sees will normally be quite messed up (AM modulation, remember?) ... and it isn't like the receiver can ask the transmitter to resend anything in a broadcast TV application. After it gets the data it is on it's own.

The second major reason that the ATSC system works as well as it does is often known by the particularly uninformative term; adaptive equalization. I will spare the reader further melodrama ("This case ... it smells like something I've seen before!"). It is, after all, time to discuss ghosts.

When a portion of the radio signal is delayed due to reflections in an analog TV system you end up with faint duplicates of the main image. In a digital AM system (ATSC) reflections cause a time delayed version of the original bit stream to be superimposed on the primary signal. This does terrible things to the receive margin. Even in the case where the reflected signal does not directly cause bit errors the resulting system will be more easily disrupted by noise. This is pretty much the last thing we want here. Ghosting is thus a huge potential problem with ATSC.

We can fix this. First we sample the signal at a rate greater than the bit rate. Then we take the series of values we end up with and look at then in a particular way. We know that there are likely several time delayed versions of the signal we want mixed in with the original signal. We also know that there are various fixed bit patterns used for synchronization that repeat on a regular basis. We then try every possible time offset and look for parts of the signal that look like these fixed bit patterns. This is a lot simpler than it sounds. You just multiply the unknown signal with the pattern you want to check for and add up the results (cross-correlation). You have to do this over many cycles and then average but fortunately the reflections in a TV application do not change very much from moment to moment. In the end you have something equivalent to a graph with time on one axis and reflection amplitude on the other which can be used to cancel the effects of the reflections. Disclaimer: Method described might not actually be used anywhere.

This works very well. An ATSC receiver is expected to be able to deal with reflections at 70% of the original signal. Other systems work even better. The DVB-T system can deal with reflections at 100% of the original signal which is pretty much as good as it is possible to be. For all practical purposes the problem of reflected signals (ghosting) no longer exists in the world of digital TV.

We now conclude that broadcast television has finally been made to work. The bigger implications of this (if any) are mostly unknown (at least to me). Some (much) smaller implications will be discussed in later posts.

posted at: 18:42 | path: /tv | permanent link to this entry