hard call to make - use mod points, or participate. I'm doing my dissertation work in an EEG lab at UCSD, and while I don't actively research neurofeedback training (NFT from hereon out), my adviser and other lab members do. The OP gets it wrong - there is no cure for these psychiatric conditions. NFT may alleviate some of the symptoms, but it is likely the underlying etiology of ADHD, PTSD and Autism (which we investigate) is different enough from individual to individual that this may not be therapeutic for everyone.
A quick crash course in EEG. Neurons in the brain communicate between themselves via chemical signals. Some of these neurotransmitters cause quick voltage changes in their target neurons (either excite with inflow of positive ions like Na+, or inhibit with inflow of negative ions like Cl-). When you have large regions of neurons communicating with others, you'll see synchronized activity - say when someone touches your arm, the part of your parietal cortex that represents that arm will have a large amount of neurons suddenly get excited. When they get excited, they draw in positive charges, and there is a net negative charge left outside of the neurons. You can detect these fluctuations in more or less real time from outside of the head - but there are limitations. Where an electrode sits on the scalp will not give you a good idea of where in the brain the signal comes from. Between the neurons and the sensor on the scalp there are several different protective layers of tissue, some fluid filled with electrolytes, bone, and skin. These electrical signals are more likely to traverse laterally underneath the skull, than to penetrate out. What you record on the outside is a noisy, noisy combination of all the signals from all over your brain, plus some muscle activity (think eye muscles and jaw). In fact, the muscle electricity is a couple orders of magnitude stronger at your scalp electrode than brain electricity. It is mathematically impossible to determine exactly where a signal recorded on the scalp originated in the brain - there are some fancy algorithms to approximate solutions, but that is another thread. Without knowing where the signals come from, one way to try to figure out what brain regions are talking to each other is by decomposing the complex signal into its component frequencies using something like a fourier transform.
Psychiatric therapy directed NFT is supposed to work by identifying different brain frequencies (from hereon out called brain rhythms) at specific locations on the scalp that differ from the general population. There is a database of the general population's brainwaves for this process called QEEG (quantitative EEG, however this is a little misleading because all EEG since digital sampling/recording is quantitative....) and you can take someone with say ADHD and compare their 10 Hz brain rhythm at the site right over the center of their head to the 10 Hz rhythm of the general population. If there is less power (power = amplitude squared) at this electrode site than the general population, a clinician might devise an NFT program to focus on the 10 hz signal at that sensor location.
p00kiethebear describes a very similar protocol for what my lab employs. For instance, we have children with a diagnosis of autism come in and watch videos or play simple games, and when the 10 Hz signal is above the threshold determined by their QEEG diagnosis, the frames will advance. It is in essence a form of guided meditation. The control that a user develops is qualitatively different from person to person. There are no clear instructions you can give someone to help them figure out how to engage a specific brain rhythm. One kid described it as imaging a hand coming out of his head. Another described peeling oranges. I seem to have a stronger 10 Hz rhythm when I imagine kung fu forms. Go figure.
Conceptually, to do this sort of training at home shouldn't be too difficult. If you have the technical savy to follow the open EEG project you'll have the minimum amount of hardware needed. Definitely read up on signal processing to figure out how to best isolate a brain rhythm of interest, and minimize muscle contamination. You may honestly have more fun reading up on some of the other brain computer interface work - like learning how to control cursors on a computer screen than just trying to increase brain rhythms in a meditative like practice. Keep in mind that to gain control of more than one dimension of commands, you'll likely need multiple electrodes and stronger signal processing skills in time x frequency space. For instance with one active electrode (at minimum you also need a second reference electrode for your differential amplifier, I recommend on the earlobe, or on the maastoid bone right behind the ear - both are electrically neutral tissue), you could learn to move a cursor up or down, but not up/down & left/right, unless you are a master of multiple brain rhythms.