Link Layer

Some images taken from http://www.cs.uni-paderborn.de/index.php?id=1119&L=1

The link layer deals with with the communication link between two neighboring nodes. In particular, the link layer is responsible for the following:

An error occurs if a frame fails to arrive at the receiver or if a frame arrives with bit errors.
The wireless medium can experience high error rates. In addition, errors are often bursty; communication will proceed without error for some time, then several errors will occur in sequence or closely together.
Approaches:

Forward error correction (FEC) - use redundancy/encoding to enable errors to be corrected at the receiver
Automatic Repeat Request (ARQ) - use ACK/NACK and retransmissions to recover lost packets or packets containing errors. Timeouts are used to detect missing packets and CRC checksums are used to detect errors.

Determine how to split network layer packets into MAC layer frames.
Each frame has overhead: MAC/PHY headers containing CRC, sender/receiver addresses, preamble for frame synchronization.
Large frames = low overhead and good bandwidth utilization
Small frames = effective error recovery since smaller chunks can be retransmitted in case of error.

PHY layer does a CRC check and discards packets with bit errors.
The MAC layer retransmits packets that are non-broadcast packets. If the sender and receiver are within radio range, communication will be reliable.
A RadioPacket (essentially MAC layer frame) is 128 bytes. The CC2420 radio has a bounded 128 byte buffer.
The Radiogram interface automatically fragments packets larger than 128 bytes.
You can bypass the link/MAC layer by using I_802_15_4_PHY to directly send packets. No ACK check will be performed.
The IProprietaryRadio provides information about the number of packets dumped because of a failed CRC check and the LowPanStats provides statistics on fragmented packets, etc.

"Datalink Streaming in Wireless Sensor Networks" by Ganti et. al.

A link layer protocol that decouples framing from error recovery.
Idea: use large frames for better bandwidth utilization but only retransmit small blocks when necessary (not entire frames)
Assumption: there will be a stream of bits coming from the network layer. Good for applications with lots of data.

A packet from the network layer is broken into blocks.
A block has 2 bytes of overhead: a sequence number and a CRC.
A series of blocks is packed into a single frame and transmitted to the MAC layer.
The sender waits for a recovery frame containing ACKs/NACks for each block. Missing blocks are retransmitted.

Blocks are of constant size. The last block may be padded with zeros to achieve this.
The last block is identified by a 1-bit in the sequence number of the block.
The last block also has a length field (1 byte) to specify the length of the data.
The maximum block sequence number is 127.
The recovery frame serves as ACK and NACK.
Recovery frame:

Type specifies that it is a recovery frame.
Src/Dest are the source and destination of the message.
Start block no. is the sequence number of the first missing or corrupt block.
The block map is a bit map where each bit indicates whether the block was received or not.

The authors looked at the potential throughput for different loss models.
The loss models varied in the size of the error clusters seen and the inter-cluster size -- the size of the blocks without error between two error clusters.

As the time between error clusters increases, the optimal block size increases.
Blocks of 20-25 bytes are optimal.
The throughput increases as the frame size increases.
The frame size used in the implementation was bounded by the packet buffer of the CC2420 radio (128 bytes).

Date: 2008-04-18