60.4.2 Data compression

An A4 scanned page of 297mm at 3.85 lines/mm is 1144 facsimile lines. This together with 1728 elements per line gives 1976832 elements per page, which if sent uncompressed to line at 9.6kbit/s would take 206 seconds, a time which can be significantly reduced by using run length coding.

60.4.2.1 Modified Huffman

With data compression a code is transmitted to represent the number of black or white bits to be sent, rather than transmitting the 1 or 0 for the appropriate scanning time. The basic G3 compression is Modified Huffman (MH) where, in a limited code table, the most frequent run lengths are represented by the shortest length codes. Because of the different distributions of black run lengths and white run lengths in typical documents, different code tables and make up codes for the runs longer than 63 elements are used. This is illus­trated in Figure 60.2 and Table 60.3, where the compression is 1728:132 or 13:1 for this particular line.

Tables 60.4 and 60.5 provides extracts from Modified Huffman code tables.

An end of line code is added to provide synchronisation and allow the receiver to check that the decoded line is 1728 elements. Typi­cally a compression of about 7:1 can be obtained.

60.4.2.2 Modified read

CCITT allows two optional 2 dimensional coding schemes. These are both RElative ADdress or READ techniques where the new line to be coded is based on the previous line. The first CCITT option of the modified version of READ (MR) was standardised before ECM (error correction mode) was available. It had to operate in the presence of network errors and ensure that the remainder of the data on the page could be decoded. To provide for this only one two dimensional line was permitted to follow the one dimensional MH line in standard resolution, and 3 lines were permitted in fine resolution. Inclusion of this redundant data in an MH line allows decoder recovery.

Figure 60.3 and Table 60.6 illustrate the Modified READ coding technique.

A pair of points either between reference and coding lines or on one or other of the lines is recognised and the appropriate code transmitted with the addition of an MH code for longer runs.

MR coding is successful because of the high degree of correlation between facsimile lines in documents and typically a compression of about 10:1 can be obtained.

When the G3 ECM option is used the reference lines in MR are redundant and CCITT permit the Modified Modified READ (MMR) option. In this case the line before the first line of the page is assumed white and all subsequent lines are referenced to their previous line. This coding is the same as T.6 used in G4 and can give compressions of the order of 14:1.

MMR is particularly suited to higher resolutions where the corre­lation between lines increases. For twice the vertical resolution with MH the number of bits to be transmitted doubles but with MMR the increase reduces to around 1.4:1.

60.4.3 Modulation and demodulation

Digital signals from the data compression encoder are unsuitable for direct transmission to the telephone network. For 9.6kbit/s the necessary transmission frequencies extend from d.c. to 4800Hz, too wide for the telephone networks nominal bandwidth of 300Hz to 3000Hz. For transmission the digital signals are transformed by the modem to analogue voice band signals which can tolerate the frequency shift, noise, amplitude and group delay variation with frequency, and also the echo and phase jitter of the international telephone network. G3 line signals are analogue. They vary in amplitude as a speech signal, but they are generated from digital signals and may be completely and accurately recovered providing line conditions permit.