Csep-561-Lec-3

Review

Shannon gives a limit for lossless data rate over a single channel with random error rate. So, we could just use more channels to improve system data rate.
You can get a decorrelated channel by just spacing out the antennas slightly.
This is MIMO (Multiple Input Multiple Output), we can quadruple rate with 4 antennas. You sense the way the channels interfere, and decorrelate them, and essentially parallelize the data rate.
This was the big differentiater in 3G -> 4G.
Diminishes returns pretty quickly, after 3 or 4 pairs of antennae.
Extends to Massive + Multi-User MIMO, where you might have a large 64x64 antenna array to serve many terminals in diverse locations at the same time.
This is a lot of 5G / Wifi 6

It's important across all layers, but in different ways at each layer.

When do we know to re-transmit?
Receiver sends an ACK frame upon receipt. If not received within some timeout, re-transmit!
Timeouts should be
not too big (cause idle), not too small (spurious resend)
On a LAN using the biggest observed ACK response works fine
On TCP/IP this is not sufficient

Generalizes stop-and-wait to allow $W$ frames to be outstanding. This is actually the state-of-the-art.
Will look at this later with TCP.

Multiplexing is the network word for the sharing of a resource
Classically, there are
TDM (Time Division Multiplexing)
Each user only gets to use channel in their time slot, round robin.
FDM (Frequency Division Multiplexing)
Multiple users simultaneously, on varying frequencies.
Static division (FDM) is useful for continuous traffic with a fixed number of users (e.g. audio streaming)
Used in telecom
TDM within FDM is used in 2G.
How do we control this? There are two classes of multiple access algorithms:
Centralized: use a privileged "scheduler" to pick who gets to transmit and when
E.g. cell towers
scales well, usually efficient
requires maangement, fairness
Distributed: all participants "figure it out"
E.g. wifi networks
operates well under low load, easy to set up, equality
scaling is really hard

A popular distributed random access protocol is MAC: Multiple Access Control.

this is the basis for classic Ethernet
The ALOHA protocol was developed in Hawaii in the 1960s to connect the Hawaiin islands.
node just sends whenever it has traffic
if there was a collision (no ACK received), then wait a random time and resend
really simple, doesn't scale well, but performs surprisingly well under low load.

This simple protocol inspired Bob Metcalfe to invent Ethernet for LANs in 1973.

Improves ALOHA by listening for activity before we send, a.k.a. **CSMA (Carrier Sense Multiple Access)
Easily done with wires, but not for wireless!
Further improved with collision detection CSMA/CD
When a conflict happens, send a JAM signal to all nodes.
To allow nodes to get the JAM signal in time (beat propagation delay):
Impose minimum frame length of 2D seconds
Ethernet minimum frame is 64 bytes
Max network length must be enforced
500m w/ coax, 100m w/ twisted pair

Much more difficult than wired

MACA (Multiple Access w/ Collision Avoidance) is a way to mitigate these problems

But with all of these solutions, there's still only one person using the medium at a time, and there are a lot of inefficiencies.

This is an alternative solution to multiple access, and is what leads to modern Ethernet.