Class 11: Error Detection

Held Wednesday, February 16, 2000

Overview

Today we continue our investigation of the data-link layer by visiting techniques for representing bits and handling potential errors in the data.

Notes

• Cool upcoming events:
  • Convocation Thursday: Richard Guy on ``Mathematics 4000 Years Ago''
  • CS Brown Bag Lunch Thursday: Growing a Language
  • Math/CS Journal Club Thursday: Fun at IBM
• When you compile your C files, it helps to use the -Wall flag (which tells the compiler to print out warnings for any odd bit of code).
  • If you use make, put the following at the top of your files and then don't explicitly specify how to make the .o files.

    CC=gcc
    CFLAGS=-Wall
    

• Are there any final questions on assignment 2?
  • I'll be in my office much of tomorrow. While I'll primarily be working on grading and some stuff I need to write, I will be happy to stop and answer reasonable questions.
• A few of you asked how I printed your code ``two up''. Here's the alias I put in my .cshrc:

    alias twoup "nenscript -2r -p- \!* | lp -oduplex";
  

• Quiz! Error correcting codes

Contents

• Data Link Issues
• Encoding Bits, Revisited
• Framing

Summary

• Encoding bit sequences
• Framing sequences

Data Link Issues

• At every level, there are some core issues we must consider.
• What are the issues we are concerned with for directly-connected nodes?
  • How do we connect the nodes?
  • How do we represent bits?
  • How do we multiplex the line?
  • How do we make chunks of bits (frames)?
  • How do we detect errors in those chunks?
  • How do we correct errors in those chunks?

Encoding Bits, Revisited

• In the previous class, we saw a few techniques for encoding bits by changing some attribute of the wave form. Let us now look at some other issues.
• We may also have to consider the clock (if clocks drift, we're going to get pretty damn confused about the signal).
• Some schemes:
  • NRZ (non-return to zero): high signal is 1, low signal is 0.
  • NRZI (non-return to zero inverted): transition is 1, non-transition is 0.
  • Manchester: exclusive or of clock and NRZ encoded data.
  • 4B/5B: encode four bits with five
• Problems with NRZ:
  • Distinguish string of 0's from lost connection
  • Distinguish high and low in a world of changing voltages (normal solution, take average of signals for midline -- what happens with long sequences).
  • What happens when clocks stray? Need encoding to help with clock recovery
• We'll do an NRZ example
• Problems with NRZI are similar
• Problems with Manchester: you need to sample the data twice as often! (Bit rate is only half of baud rate.)
• 4B/5B gives a compromise. By using an extra bit, you can ensure that there is sufficient change across time.

  4-bit	5-bit		4-bit	5-bit
  0000	11110		0001	01001
  0010	10100		0011	10101
  0100	01010		0101	01011
  0110	01110		0111	01111
  1000	10010		1001	10011
  1010	10110		1011	10111
  1100	11010		1101	11011
  1110	11100		1111	11101

• No, I have no clue how they came up with these (other than that the last bit gives parity). Do you?

Framing

• It is helpful to break data into small chunks, which we typically call ``frames''.
• Why?
  • By grouping information, we can more easily insert error detecting codes and error correcting codes.
  • By grouping information, we can ``name'' chunks, which is useful if we need to request retransmission
• Most framing mechanisms use some special values to designate the start (and, often, the end) of the frame.
• Most framing mechanisms include some other information (the frame header) at the start of the frame.
• What do we do if the special ``start of frame'' values appear within the body of the frame?
• Byte stuffing solution: Prefix those values with a special byte, like the backslash in C.
  • Advantages: easy; fairly reliable
  • Disadvantages: adds overhead; frame length differs; a bit error in the ``quote byte'' can be disasterous
• Bit stuffing solution: Use special sequences and insert an extra bit if those sequences appear.
  • HDLC (High-level Data Link Control) uses 01111110
  • After five consecutive 1's in the body, it inserts a 0.
  • The receiver knows that 01111110 is the end of the body.
  • 01111111 is an error
  • 01111101 is really 0111111
  • Advantages: easy, fairly reliable
  • Disadvantages: frames do not have a uniform length; two bit errors can lead to a fake end-of-frame
• Clock based framing solution: The start-of-frame value appears at a specified interval in the data. If it appears out of synch, it must be part of the data.
  • Advantages: no overhead
  • Disadvantages: recovery may require multiple frames