Now the problem is that it is despite those two great ideas, you can run into

the following situation. Let's say you are station A, okay, and

you want to say talk to station B, which is an access point, and then someone else

across from you in the Starbucks or in a hotel room hotel lobby want to talk to

this same AP. Now, if you both talk the same time

actually, you will collide, okay. But, let's say, your sensing range has

only this radius, which is the same as your, let's say, transmission range,

okay. So, you can not sense the existence of C,

and when C is talking actually you can't even hear it.

So you may just start talking again, and then be, suddenly get confused.

So this is the problem that, two stations in WiFi may interfere.

But they do not sense each other. And this is one of the fundamental

limitation that says that if you want to use random access,

and yet, sensing range is not the same as interference range,

then you have a problem. So how can we tackle this problem?

Now, so far, we have used no explicit message passing yet.

Now we're going to need a little explicit message passing, have to send some

control signals, what we call rts cts, okay, request to send and clear to send.

So the basic idea is that, the sender. Okay, when you want to send something you

first send a control packet. It's a tiny one called a request to send,

'kay? And then, upon a small listening

interval, the receiver will send another control packet called, clear to send.

And then after another little interval, you can start sending.

Now, whys would this work? It works, because when you send a request

to send. Okay.

Everybody in your int, transmission range hears this.

So B hears this, and then B will then send a clear to send.

And everybody that B can talk to can hear this, including yourself and others who

might want to talk to B. Oaky.

Now assuming a symmetric, transmission range.

B talk to C is same to same as C talk to B for now.

Now, C will then say, Oh! I didn't ask to send something, but I get

a clear tone signal, that means somebody else is going to send something.

So, I will refrain from talking. Okay again all these discussion here on

tcp assume that the network elements obey the protocols.

If they actually try to trick the system then you need some other kind of game

thematic analysis. Alright so now A says hey I send RTS and

I got a CTS. That means I'm clear to send because all

potential interfere I cannot hear have got this CTS and will they refrain from

sending. So, that's the smart idea of RTS-CTS use

a little sequence. Okay?

So, there's the overhead. You have to pay the overhead of the

spending this much time just doing control signal, signalling in order to

guaranty that the data will not be colliding, even when you cannot sense

somebody. So, this is the famous hidden node

problem. C send a note to a or vice versa and, rts

cts solution of that problem. Of course it's not perfect, for example,

the rts packets may collide, so it does not completely solve the issue, but helps

quite a bit. All right now, we're going to try to

understand csma built upon the ideas one, two, three that we just mentioned.

Okay, wait and listen through carrier sensing,

Use randomize the binary exponential backup upon collision, that is, no

acknowledgement back, and use RTS CTS explicit control messaging if needed.

Well it turns out that analyzing a CSMA and WiFi is not that easy, okay?

For a more proper understanding we need to use something called two dimensional

mark of chain to specify the protocols, because there's a lot of probabilistic

actions and extracting those out like in TCP our discussion of TCP will not have

enough predictive power. So we need to do some probability theory

and instead of doing a two dimensional mark of chain which we will not have the

prerequesites in this class. We will do a very basic, a little hand

wavy type of basic probability argument. The main difficulty here is that, the

frame collision. Depends on the action of each radial, as

well as the history of binary exponential back off.

So put B. And the binoexponential backoff is turn

coupled with the transmission decision. Should I transmit or not which is again

probabilistic by each station. So because of this complication even this

hand wavy simplified version of argument would take a little derivation, okay,

through basic probability theory. Now the basic idea where we want to do is

we want to look at true probabilities. One is the probability, the probability

okay, of transmission. At the given time slot.

The other is the probability of collision.

And then we'll try to relay to these two probabilities, and derive two equations,

and solve two variables. Now let's go into the detail here.

First, a little notation, say we want to derive the performance metric of CSMA

random access. Denote that by S, okay, which is really

the average through put, it's the average number.

Of bits, transmitted successfully over. Successfully transmitted.

Okay. A time slot divided by the average length

of a time slot. There're different kinds of time slots

and this is an average notion. Okay so average number of bids

successfully transmitted over the average length of a time slot.

That is as they will want to arrive, and want to see how does s depends on

different barometers in problem. Okay, we're going to use a p sub t to

denote the probability. Of, transmission.

In fact, I should say a probability of at least one, it could be more transmission

in the entire wifi network. And then we use P sub S to denote the

probability that the transmission is successful.