1. Video Conferencing System
EECS150 Spring 2008
Shah Bawany
9/16/2018
EECS150 Spring Shah Bawany

2. Checkpoint Overview (1)
PIP Display
Text Interface
DCT (Blackboxed)
Huffman Coder
Packetizer/Depacketizer
Transmission Protocol
Extra Credit
9/16/2018
EECS150 Spring Shah Bawany

3. Checkpoint Overview (2)
Transceiver
Transmission Protocol FSM
Packet
Buffer
Packet
Buffer
Huffman Coder
Huffman Decoder
DCT
IDCT
From CP2.5
To CP2.5
9/16/2018
EECS150 Spring Shah Bawany

4. EECS150 Spring 2008 - Shah Bawany
Checkpoint 2.5
Checkpoint 4 requires you to finish all the requirements for Checkpoint 2.5 including:
Text display at the bottom of the screen
Picture-in-picture display
Subsampled local and remote video stored in SDRAM
Modified address counters for wireless video processors
9/16/2018
EECS150 Spring Shah Bawany

5. Discrete Cosine Transform
In general the DCT is a multi-dimensional transform much like the Fourier Transform
We will be using a 2-D transform on 8 x 8 subimages from our compressed frame
It generates real-valued spatial frequencies in a new 8 x 8 matrix
Why we do it?
Images are generally low frequency, so many of the frequency components will zero out
It concentrates much of the information into values clustered in to the upper left corner of each matrix
This gives us a more tightly clustered distribution allowing us to exploit variable length coding as an added compression technique
9/16/2018
EECS150 Spring Shah Bawany

6. EECS150 Spring 2008 - Shah Bawany
Quantization
Since the frequency component coefficients generated by the DCT span such a large range, relatively small changes to the lower order bits don’t make much of a difference on the appearance of the resulting image
We can take the real-valued (signed fixed point) DCT coefficients and simply truncate them to integer multiples of 8 to save some space
Greater quantization means less space is used and more information is lost.
We have determined a suitable threshold at which the appearance degrades significantly
9/16/2018
EECS150 Spring Shah Bawany

7. Prefix-Free Variable Length Coding
When we have a source where some symbols are more frequent than others, it can be beneficial to convert each of the symbols into a unique compressed codeword with length inversely proportional to its frequency of occurrence
For instance if we have a collection of 2-bit values composed of 1/8 0’s, 1/8 1’s, 1/8 2’s and 5/8 3’s.
Uncompressed expected length: 2 bits
Compressed using the scheme:
3 = 0
2 = 10
1 = 110
0 = 111
5/8 + 2(1/8) + 3(1/4) = 1 5/8 < 2… Success!
Why those specific codewords?
We want to make sure they’re prefix free
9/16/2018
EECS150 Spring Shah Bawany

8. Prefix-Free Variable Length Coding (2)
Prefix free means that if we took the codewords as a bitstream, we would know when we have finished reading a single codeword
For instance, using our previous coding scheme the bitstream will be decomposed as:0 -> 3
0 -> 3
10 -> 2
111 -> 0
etc
You can see that there is no ambiguity because once we have looked at the first 0, there are no other codewords that start with 0
9/16/2018
EECS150 Spring Shah Bawany

9. Prefix-Free Variable Length Coding (3)
More formally, for all X in our code alphabet C, there exist no codewords Y in C such that Y is a prefix of X
Luckily for you, we have provided a static coding scheme that is based on frequency of occurrence of values in our DCT coefficient matrices
We will be using signed unary-binary code, which works as follows
0 maps to 1
All other values N map to
{log2N ‘b0, |N |, (N<0) ? 1’b1 : 1’b0}
For example 5 would be , while -12 would be
Your coder will accept parallel inputs and output a bitstream serially
9/16/2018
EECS150 Spring Shah Bawany

10. Packetizing / Depacketizing
The output of the Huffman Coder is a bitstream and we wish to send packets containing 31 bytes of raw video data
You should completely fill each data packet, ignoring where symbols start and end
Assuming that you don’t lose any packets, the bitstream can be completely recovered and since the code is prefix-free, there is no need for delimiters or special grouping
You can accomplish this with shift registers and FIFOs
9/16/2018
EECS150 Spring Shah Bawany

11. Transmission Protocol
Our protocol this semester will contain the following packet formats:
SYN: Master’s request to establish a connection
SYNACK: Slave’s acknowledgement / acceptance of connection attempt
DATA: compressed video stream data
To ensure that the protocol is lossless we must retransmit if a packet is lost
To simplify the protocol and reduce overhead, we simply send a packet and wait for receipt of any packet in response as an acknowledgement of the first packet
In the event of failure, wait a specified amount of time and retransmit
If successful, continue transmitting
9/16/2018
EECS150 Spring Shah Bawany

12. Transmission Protocol (2)
Transmitting fast increases the probability of error, so there’s a tradeoff between decreasing retransmissions and increasing the rate of transmission
For the sake of Checkpoint 4, you can find a reasonable transmission rate (once every 4ms like CP3 fast transmit)
For extra credit you can have your solution speed up or slow down dynamically based on channel congestion by optimizing the equation:
[(1/(1-p)) - 1]Tretransmit + Ttransmit
9/16/2018
EECS150 Spring Shah Bawany

13. Transmission Protocol (3)
Timeout
Master Reset
SYN
SYNACK
Clock
Slave Reset
DATA
Timeout
Slave
Wait
SYN
SYN
ACK
DATA
GET
DATA
DATA
Timeout
9/16/2018
EECS150 Spring Shah Bawany

14. Transmission Protocol (4)
Master
Slave
SYN
SYNACK
DATA0
DATA1
DATA 1[Lost]
(wait for retransmit)
DATA1 [re-TX]
DATA2 (serves as ACK)
DATA2
9/16/2018
EECS150 Spring Shah Bawany

15. Transmission Protocol (5)
MPDU
Src Dst WL Payload
Header
Video Data
1-Byte header
31-Byte bistream
Packet Type
Seq ID
7:3 2:0
9/16/2018
EECS150 Spring Shah Bawany

16. A Sample Scenario 9/16/2018 EECS150 Spring 2008 - Shah Bawany Subimage31 25 10 9 30 20 18 23 4 2 1 98 97 22 5 3
DCT
Original Image 99 -5 10 4 20 18 7 -9 2 12 1 2 -1
Quant.
Output from the Huffman Coder:
146 bits from an original 448!
(Note this is not a real DCT result, but it simulates actual behavior)
9/16/2018
EECS150 Spring Shah Bawany