The IIS Chip Gallery
Arithmetic_Coder (1993)
Additional pictures below, click to see larger versions
by
| Application | Communication |
| Technology | 1500 |
| Manufacturer | VLSI Tech |
| Type | Semester Thesis |
| Package | DIP40 |
| Dimensions | 3600μm x 3800μm |
| Gates | 3 kGE |
| Voltage | 5 V |
| Clock | 8 MHz |
The goal of this student project was to design an ASIC for lossless compression/decompression of black and white images. Huffman coding, a source coding scheme that assigns short codes to frequent data symbols and longer codes to less frequent symbols, is ineffective on two-valued data because there is no code word shorter than a single bit.
Arithmetic data compression can be viewed as an extension of Huffman coding that overcomes this limitation. The basic idea is to represent an entire message by a real number obtained from successively narrowing down the number interval [0,1] from both sides as a function of the symbols found and of a statistical model of their source.
Our design uses context-dependent frequencies of occurence that are adjusted once per frame. The context taken into account consists ofa total of six pixels from the actual scanning line and from the previous one. For reasons of economy, previous line data are stired in a separate FIFO. On the average, a compression ratio of 20 is obtained on telefax data.