GSM: Radio interface

One of the main objectives of GSM is roaming. Thus, to allow for interoperability between MNs stations and disparate networks of the radio interface must be standardised. Spectrum efficiency depends on aspects of the radio interface and transmission, such as system capacity or techniques used to optimize SIR and frequency reuse. It thus, becomes clear that the specification of the radio interface can influence the spectrum efficiency.

Frequency allocation
Two frequency bands, of 25 Mhz each, are allocated for the GSM system:
y y

890-915 Mhz for the uplink (MN to BTS). 935-960 Mhz for the downlink (BTS to M).

However, for reasons related to the military as well as the existence of past analog systes (that use part of the two frequency bands), not all the countries can use the whole GSM frequency bands.

Medium access
GSM employs a mix of Frequency Division Multiple Access (FDMA) and Time Division Multiple Access (TDMA), combined with frequency hopping. Using FDMA, a frequency is assigned to each user. So for large number of users in a FDMA system, the larger the number of required frequencies. The limited available radio spectrum and the fact that a user will not free its assigned frequency until he does not need it anymore, reasons about scalability problems in an FDMA system. TDMA allows several users to share the same channel. Each subscriber multiplexes the shared channel, scheduling their frame for transmission. Usually TDMA is used with an FDMA structure. In GSM, a 25 Mhz frequency band is divided, using a FDMA scheme, into 124 carrier frequencies with a 200khz spacing. Normally a 25 Mhz frequency band can provide 125 carrier frequencies; however, the first carrier frequency is used as a guard-band between GSM and other services working on lower freq. band. Each carrier is time-divided using a TDMA scheme. This scheme splits a 200khz channel, into 8 bursts. A burst is the unit of time in a TDMA system, and it lasts approximately 0.577ms. Thus a TDMA lasts 4.615ms. Each burst is assigned to a single user.

Channel structure

A channel maps to the recurrence of one burst every frame. It is defined by its frequency and the position of its corresponding burst within a TDMA frame. In GSM there are two types of channels:
y y

traffic channels used for speech and data. control channels used for network management messages and channel maintenance tasks. Traffic channels (TCH)

Full-rate traffic channels (TCH/F) are defined using a group of 26 TDMA frames called a 26Multiframe. The 26-Multiframe lasts 120 ms. In this frame group traffic channels for the downlink and uplink are separated by 3 bursts. That implies, the mobiles will not need to transmit and receive at the same time which simplifies considerably the electronics of the system. The frames that form the 26-Multiframe structure have different functions:
y y y

24 frames are reserved to traffic. 1 frame is used for the Slow Associated Control Channel (SACCH). The last frame is unused. It allows the MN to perform other functions, such as measuring the signal strength of neighboring cells.

Half-rate traffic channels (TCH/H), which double the capacity of the system, are also grouped in a 26-Multiframe but the internal structure is different.

Control channels
According to their functions, 4 different classes of control channels are defined: Broadcast channels (BCH) The BCH channels are used, by BTS to provide the MN with synchronization information from the network. 3 different types of BCHs can be distinguished:
y y y

Broadcast Control Channel (BCCH): gives to the MN the parameters needed to identify and access the network. Synchronization Channel (SCH): gives the MN the training symbol sequence to demodulate the information transmitted by BTS. Frequency-Correction Channel (FCCH): provides the MN with the frequency reference of the system for the purposes of syncronisation. Common Control Channels (CCCH)

The CCCH channels help to establish the calls from the mobile station or the network. These are:
y y y

Paging Channel (PCH): used to alert the MN of an incoming call. Random Access Channel (RACH): used by the MN to request network access. Access Grant Channel (AGCH): used, by the BTS, to inform the MN about the channel it should use. This channel is the answer of a BTS to a RACH request from the MN. Dedicated Control Channels (DCCH)

The DCCH channels are used for message exchange between several mobiles or a mobile and the network. These are:
y y

Standalone Dedicated Control Channel (SDCCH): used to exchange signaling in the downlink and uplink. Slow Associated Control Channel (SACCH): used for channel maintenance and control. Associated Control Channels (ACCH)

Fast Associated Control Channels (FACCH) replace all or part of a traffic channel when urgent signaling must be transmitted. The FACCH channels carry the same signaling as SDCCH channels.

Burst structure Four different types of bursts can be distinguished in GSM:

y y y y

Frequency-correction, used on the FCCH. It has the same length as the normal one but a different structure. Synchronization burst used on the SCH. It has the same length as the normal one but a different structure. Random access used on the RACH and is shorter than the normal burst. Normal burst used to carry speech or data information. It lasts approximately 0.577 ms and has a length of 156.25 bits. Its structure is presented below.

Structure of the 26-Multiframe, the TDMA frame and the normal burst The tail bits (T) are a group of 3 bits set to zero and placed at the beginning and the end of a burst. They cover the periods of ramping up and down of the mobile's power. The coded data bits corresponds to two groups, of 57 bits each, containing signaling or user data. The stealing flags (S) indicate, to the receiver, whether the data bits are data or signaling traffic. The training sequence has a length of 26 bits. It synchronizes the receiver, thus masking out multipath propagation effects. The guard period (GP), with a length of 8.25 bits, is used to avoid a possible overlap of two mobiles during the ramping time. Frequency hopping Propagation effects and thus, multipath fading depend on the radio frequency. To eliminate significant differences in channel quality, slow frequency hopping is introduced; it changes the frequency with every TDMA frame (fast frequency hopping changes the frequency many times per frame but it is not used in GSM). The frequency hopping also reduces the effects of cochannel interference. There are different types of frequency hopping algorithms. The algorithm selected is sent through BCCH. Frequency hoping is optional for a BTS but must be supported by the MN.

From bits to radio

The following figure shows the steps involved to transform speech audio to radio waves and vice versa.

From bits to radio

If the source of information is data (not speech), the speech coding is not performed.

Speech coding Talkspurts transmission (voice audio) is the mainstream service of a cellular system. The GSM speech codec that transforms the analog signal (voice) into a digital representation, must meet the following criteria:
y y

Maintain speech quality at least equal to previous cellular systems. Reduce redundancy in voice utterances. This reduction is essential due transmission capacity limitation on the data channel.

Adopt low complexity speech codec to reduce production costs.

The standard GSM speech codec is RPE-LTP (Regular Pulse Excitation Long-Term Prediction). This codec uses statistics from previous samples (information that doesn't change very quickly) to predict the current sample. The speech signal is divided into blocks of 20ms. These blocks are then passed to the speech codec of 13 kbps, to obtain sppech frames of 260 bits each.

Channel coding Channel coding adds redundancy bits to the original information to detect and correct, if possible, transmission errors. Channel coding for the GSM data TCH channels Channel coding is performed using two codes: a block and a convolutional code. The block code is defined in the GSM Recommendations 05.03. It receives an input block of 240 bits and adds 4-zero tail bits at the end of the input block; this results a block output of 244 bits. A convolutional code adds redundancy bits to protect the data. A convolutional encoder contains memory. This property differentiates the two types of code. A convolutional code can be defined by three variables : n, k and K. The value n corresponds to the number of output bits from the encoder, k to the number of input bits and K to the memory of the encoder. The ratio (R) of the code is defined as R = k/n. For example, a convolutional code with k=1, n=2 and K=5, uses a ratio of R = 1/2 and delay of K=5, which means that it will add oneredundant bit for each input bit (1 in 2 output bits is an input bit). The code uses 5 consecutive bits to compute the redundancy bit. As the convolutional code is a 1/2 rate for an input block of 244 bits an output block of 488 bits is generated. These 488 bits are punctured to produce a block of 456 bits. 32 bits, obtained as follows, are not transmitted : C (11 + 15 j) for j = 0, 1, ..., 31 The output block of 456 bits is then passed to the interleaver.

Channel coding for the GSM speech channels Before applying channel coding, the 260 bits of a GSM speech frame are divided in 3 different classes according to function and importance. The most important class is the class Ia containing 50 bits. Next in importance is the class Ib, which contains 132 bits. The least important is the class II, which contains the remaining 78 bits. The different classes are coded differently:

y y

y y

Class Ia bits are block-coded. 3 parity bits are added to the 50 class-Ia bits. The Ia output (53 bits) are added to the Class Ib bits (50+3+132); 4 zero bits are added to the Ia+Ib bits (185+4). A convolutional code, with r = 1/2 and K = 5, is then applied, obtaining an output block of 378 bits (189*2). Class II bits are added (378+78), without any protection, to the output block of the convolutional coder. The 456-bit block is finally constructed.

Channel coding for the GSM control channels In GSM, signalling information is contained in just 184 bits. 40 bits parity, obtained using a fire code, and 4-zero bits are added to the 184 bits before applying the convolutional code (r = 1/2 and K = 5). The output of the convolutional code is then a block of 456 bits; it does not need to be punctured.

Interleaving
This method rearranges a group of bits in a particular way. It is combined with FEC codes in order to improve the performance of the error correction mechanisms. Interleaving decreases the possibility of losing whole bursts during the transmission, by dispersing the errors. Since the errors become less concentrated, it is then easier to correct them.

Interleaving for the GSM control channels At the physical layer a burst in GSM transmits 2 blocks of 57 data bits each. Thus, the 456-bit block output of the channel coder fit into 4 bursts (4*114 = 456). The 456 bits are, thus, divided into 8*57-bit blocks. As interleaving is applied during the forming of the blocks, the 1st block of 57 bits contains the bit numbers (0, 8, 16, .....448), the second one the bit numbers (1, 9, 17, .....449), etc. The last block of 57 bits will then contain the bit numbers (7, 15, .....455). The first 4 *57-bit blocks are placed in the even-numbered bits of four bursts. The other 4 are placed in the odd-numbered bits of the same four bursts. Therefore, the interleaving depth of GSM interleaving for control channels is 4 and a new data block starts every 4 bursts. The interleaver for control channels is called a block rectangular interleaver.

Interleaving for the GSM speech channels The 456-bit block, obtained after the channel coding, is divided in 8*57-bit blocks in the same way as it is explained in the previous paragraph. But these 8 blocks are distributed differently.

The first 4 blocks of 57 bits are placed in the even-numbered bits of 4 consecutive bursts. The other four blocks are placed in the odd-numbered bits of the next four bursts. The interleaving depth of the GSM interleaving for speech channels is then 8. A new data block also starts every 4 bursts. The interleaver for speech channels is called a block-diagonal interleaver.

Interleaving for the GSM data TCH channels A particular interleaving scheme, with an interleaving depth equal to 22, is applied to the block of 456 bits obtained after the channel coding. The block is divided into 2 blocks of 6 bits each, 2 blocks of 12 bits each, 2 blocks of 18 bits each and 16 blocks of 24 bits each. It is spread over 22 bursts in the following way :
y y y y

the 1st and 22nd bursts carry one block of 6 bits each (2) the 2nd and 21st bursts carry one block of 12 bits each (2) the 3rd and 20th bursts carry one block of 18 bits each (2) from 4th to 19th burst, a block of 24 bits is placed in each burst (16)

A burst will then carry information from 5-6 consecutive data blocks. The data blocks are said to be interleaved diagonally. A new data block starts every four bursts.

Burst assembling The burst assembling step manages the grouping the bits into bursts.

Encryption It is used to protect signaling and data. An encryption key is computed using: 1. algorithm A8 (stored on the SIM card), 2. the subscriber key 3. a random number (nonce) delivered by the network (same as the one used for authentication).