The acronym LTDMA stands
for Learning-automata Time Division Multiple Access.
LTDMA is a MAC sublayer
protocol which belongs to the TDMA protocol family. Protocols of
the TDMA family (like RTDMA) divide time in timeslots
and each station can only transmit a packet in such a timeslot. TDMA
uses a round-robin station selection scheme, i.e. all stations
transmit their packets the one after the other.
The problem with that scheme, is that a
vast percentage of timeslots is wasted because computer networks
produce bursty traffic as Tanenbaum suggests in the bible of
computer networks... "Computer Networks". That means that while some
stations may have large amounts of traffic to transmit and thus need
bandwidth, other stations may be idle but nevertheless wasting valuable
bandwidth. To correct this problem, LTDMA does not use a blind
round-robin station selection scheme but instead, it uses a learning
automaton at each station which keeps the choice probability of all
stations. Let me explain this a bit further.
The whole point is to keep track of
which stations transmit packets and which remain idle, so that
timeslots can be allocated to the stations that really need them. In
order to achieve that, everytime a station is selected by the protocol
(we will discuss how this is done), its choice probability is decreased
in case the station does not have a packet to transmit, whereas the
protocol increases the station's choice probability if the station
successfully transmitted a packet. This way, stations which remain idle
are assigned a very low choice probability and therefore do not waste
further timeslots, whereas busy stations (such as ftp servers, etc)
have a high choice probability and are thus very frequently selected.
Below follow the equations which control the choice probabilities:
Pi <= Pi + L(1 - Pi)
Pi <= Pi - L(Pi -
a)
Where L, a ε (0,1)
Pi is the choice
probability of station i.
L is a factor that has a value near 1 (about 0.90) and the
closer to 1, the quicker the station probability convergence.
a is a near zero parameter that prevents the choice probability of a station to drop to zero, in case of many unsuccessfull transmits.
LTDMA is a fully decentralized protocol which requires no central coordination of the stations. Now, the question is: How do the individual stations know who's turn it is to transmit?
The answer is pretty easy. Every station has the learning automaton that we mentioned above, which contains the choice probabilities of all other stations. That means that all stations essentially contain a copy of exactly the same automaton. This is made possible by the broadcast nature of the medium, which allows all stations to receive the same feedback information (if a station transmitted a packet or not) and update their automata accordingly.
Now, all stations using a common pseudorandom number generator seeded with a common initial value, produce before each timeslot a pseudorandom number (all stations the same number of cource), according to which the station which will transmit is selected. That way, at each timeslot, all stations select according to the choice probabilities the same station. Therefore, although LTDMA requires absolutely no central coordination, the protocol is collision-free and achieves a high performance even in the highly bursty traffic conditions of a computer network.