GSM TRANSMISSION PROCESS

STAGE 1: ANALOG TO DIGITAL (A/D) CONVERSION

One of the primary functions of an MS is to convert the analog speech information into digital form for transmission using a digital signal. The analog to digital (A/D) conversion process outputs a collection of bits: binary ones and zeros which represent the speech input.

A/D conversion

The A/D conversion is performed by using a process called Pulse Code Modulation (PCM). PCM involves three main steps:

• Sampling

• Quantization

• Coding

Step 1: Sampling

Sampling involves measuring the analog signal at specific time intervals.

Analog signal sampling

The accuracy of describing the analog signal in digital terms depends on how often the analog signal is sampled. This is expressed as the sampling frequency. The sampling theory states that: To reproduce an analog signal without distortion, the signal must be sampled with at least twice the frequency of the highest frequency component in the analog signal. Normal speech mainly contains frequency components lower than 3400 Hz. Higher components have low energy and may be omitted without affecting the speech quality much. Applying the sampling theory to analog speech signals, the sampling frequency, should be at least 2 x 3.4 kHz = 6.8 kHz. Telecommunication systems use a sampling frequency of 8 kHz, which is acceptable based on the sampling theory.

Step 2: Quantization

The next step is to give each sample a value. For this reason, the amplitude of the signal at the time of sampling is measured and approximated to one of a finite set of values. The figure below shows the principle of quantization applied to an analog signal. It can be seen that a slight error is introduced in this process when the signal is quantized or approximated. The degree of accuracy depends on the number of quantization levels used. Within common telephony, 256 levels are used while in GSM 8,192 levels are used.

Quantization

Step 3: Coding

Coding involves converting the quantized values into binary. Every value is represented by a binary code of 13 bits (2¹³=8192). For example, a quantized value of 2,157 would have a bit pattern of 0100001101101:

Coding of quantised value 2157

Summary of A/D Conversion

The result from the process of A/D conversion is 8,000 samples per second of 13 bits each. This is a bit rate of 104 kbits/s. When it is considered that 8 subscribers use one radio channel, the overall bit rate would be 8 x 104 kbits/s = 832 kbits/s. Recalling the general rule of 1 bit per Hertz, this bit rate would not fit into the 200 kHz available for all 8 subscribers. The bit rate must be reduced somehow - this is achieved using segmentation and speech coding.

STAGE 2: SEGMENTATION AND STAGE 3: SPEECH CODING

The key to reducing the bit rate is to send information about the speech instead of the speech itself. This can be explained with the following analogy:

Person A wishes to listen to a certain piece of music and they know that person B has it on record. A rings B asking for the use of the record for some time. Unfortunately, the record is scratched and cannot be used. Instead, B sends A parameters of how the music is built up - the sheets of music - together with information about how fast it should be played - the frequency - and A reproduces the music. In GSM, the speech coding process analyzes speech samples and outputs parameters of what the speech consists of the tone, length of tone, pitch, etc. This is then transmitted through the network to another MS, which generates the speech based on these parameters. The process of segmentation and speech coding is explained in more detail as follows:

The human speech process starts in the vocal chords or speech organs, where a tone is generated. The mouth, tongue, teeth, etc. act as a filter, changing the nature of this tone. The aim of speech coding in GSM is to send only information about the original tone itself and about the filter.

Segmentation: Given that the speech organs are relatively slow in adapting to changes, the filter parameters representing the speech organs are approximately constant during 20 ms. For this reason, when coding speech in GSM, a block of 20 ms is coded into one set of bits. In effect, it is similar to sampling speech at a rate of 50 times per second instead of the 8,000 used by A/D conversion.

Segmentation and speech coding

Speech Coding: Instead of using 13 bits per sample as in A/D conversion, GSM speech coding uses 260 bits. This calculates as 50 x 260 = 13 kbits/s. This provides a speech quality which is acceptable for mobile telephony and comparable with wireline PSTN phones. Many types of speech coders are available. Some offer better speech quality, at the expense of a higher bit rate (waveform coders). Others use lower bit rates, at the expense of lower speech quality (vocoders). The hybrid coder which GSM uses provides good speech quality with a relatively low bit rate, at the expense of speech coder complexity.

Speech quality vs. bit rate

Summary of Segmentation and Speech Coding

The GSM speech coder produces a bit rate of 13 kbits/s per subscriber. When it is considered that 8 subscribers use one radio channel, the overall bit rate would be 8 x 13 kbits/s = 104 kbits/s. This compares favorably with the 832 kbits/s from A/D conversion. However, speech coding does not consider the problems which may be encountered on the radio transmission path. The next stages in the transmission process, channel coding and interleaving, help to overcome these problems.

STAGE 4: CHANNEL CODING

Channel coding in GSM uses the 260 bits from speech coding as input to channel coding and outputs 456 encoded bits. The 260 bits are split according to their relative importance:

• Block 1: 50 very important bits

• Block 2: 132 important bits and

• Block 3: 78 not so important bits

The first block of 50 bits is sent through a block coder, which adds three parity bits that will result in 53 bits. These three bits are used to detect errors in a received message. The 53 bits from first block, the 132 bits from the second block and 4 tail bits (total = 189) are sent to a 1:2 convolutional coder which outputs 378 bits. Bits added by the convolutional coder enable the correction of errors when the message is received. The bits of block 3 are not protected.

Channel coding

STAGE 5: INTERLEAVING

First level of interleaving

The channel coder provides 456 bits for every 20 ms of speech. These are interleaved, forming eight blocks of 57 bits each, as shown in the figure below.

Interleaving of 20 ms of encoded speech

In a normal burst there is space for two of these speech blocks of 57 bits. (The remaining bits in the burst are explained later in this book.) Thus, if one burst transmission is lost, there is a 25% BER for the entire 20 ms of speech (2/8 = 25%).

Normal burst

Second level of interleaving

If only one level of interleaving is used, a loss of this burst results in a total loss of 25%. This is too much for the channel decoder to correct. A second level of interleaving can be introduced to further reduce the possible BER to 12.5%. Instead of sending two blocks of 57 bits from the same 20 ms of speech within one burst, a block from one 20 ms and a block from next sample of 20 ms are sent together. A delay is introduced in the system when the MS must wait for the next 20 ms of speech. However, the system can now afford to loose a whole burst, out of eight, as the loss is only 12.5% of the total bits from each 20ms speech frame. 12,5% is the maximum loss level that channel decoder can correct.

Speech frame

Second level of interleaving

STAGE 6: CIPHERING/ENCRYPTION

The purpose of ciphering is to encode the burst so that it cannot be interpreted by any other device than the intended receiver. The ciphering algorithm in GSM is called the A5 algorithm. It does not add bits to the burst, meaning that the input and output to the ciphering process is the same as the input: 456 bits per 20 ms.

STAGE 7: BURST FORMATTING

As previously explained, every transmission from an MS/BTS must include some extra information such as the 26 training sequence bits, 2 flag bits and 6 tail bits. The process of burst formatting is to add these bits to the basic speech/data (57+57=114 bits) being sent. Consequently this increases the burst from 114 to 148 bits, thus increasing the transmission rate on the air, but is necessary to counteract problems encountered on the radio path. In GSM, the input to burst formatting for 20ms speech is the 456 bits received from ciphering. Burst formatting adds 136 bits to it, bringing the sum total to 592. However, each time slot on a TDMA frame is 0.577 ms long. This provides enough time for 156.25 bits to be transmitted (each bit takes 3.7 μs), but a burst only contains 148 bits. The rest of the space, 8.25 bit times, is empty and is called the Guard Period (GP). This time is used to enable the MS/BTS “ramp up” and “ramp down”. To ramp up means to get power from the battery/power supply for transmission. Ramping down is performed after each transmission to ensure that the MS is not transmitting during time slots allocated to other MS’s. The output of burst formatting is a burst of 156.25 bits(one burst) or 625 bits(four bursts) for 20 ms sample. Thetransmission bit rate for GSM can be calculated to be 270.9 kbits/s(156.25/.577).

STAGE 8: MODULATION & TRANSMISSION

The bits must then be sent over the air using a carrier frequency. As previously explained, GSM uses the GMSK modulation technique. The bits are modulated onto a carrier frequency and transmitted (e.g. 912.2 MHz). The following figure summarizes the GSM transmission process.

GSM transmission process