Audio is sampled at 8 kHz rate, and divided into 20 ms blocks of 160 samples per block. Each 20 ms block is characterised by 260 bits i.e. 13 kbits/sec. The resulting signal is processed by a regular pulse excitation - long term predictor (RPE - LTP) codec. This yields a digital representation of vocal chord vibrations, together with the filter characteristics which must be applied to them to make voice sounds. The most significant 182 bits will go through error correction and become 378 bits. The less critical group of 78 bits will not go through error correction and will just be summed with the 378 bits which will yield 456 bits.
These 456 bits are then separated in 8 blocks of 57 data bits. These blocks are interleaved with adjacent blocks to guard against burst errors and broken up into blocks of 114 bits for transmission. This block of 114 bits are the data bits of the timeslot. Timing data is added, and the resulting bit stream is fed to the Gaussian Minimum Shift Keying (GMSK) modulator, where the bits are taken two at a time and used to smoothly change the phase of an RF carrier according to bit combination.
4.1
Full rate / Half rate / Enhanced full rate.
The data rate of 13 kbit/sec ( full rate ) is considerably lower than for direct speech digitising as in PCM. Now more advanced voice coders cut this to 5.6 kbit/sec ( half rate coding ). The enhanced full rate, is just a full rate with a different speech coder which improves the transmission quality. Mitsubishi Electric Telecom Europe SA ZA le Piquet, 35370 Etrelles Phone: +33 2 99 75 71 00 Fax: + 33 2 99 75 71 47 Version E Date: 04/00
