Global system for mobile communications (GSM)

Following the standardization and the global success of the Pan-European digital mobile cellular radio system it was termed as the Global System of Mobile communications, or GSM in short. In June 2006 the mobile phone industry celebrated the historic moment of connecting the second billionth GSM mobile phone user on the globe. According to the GSM Association (GSMA), 1000 new users were signing up per minute, when including both second generation GSM, as well as to so-called third generation 3GSM services. Technically speaking, this corresponds to a 'subscription rate of about 18Hz' - it 18 users/sec. In order the satisfy the demand of both new and existing customers, 'new mobile phone are rolling off the global productions line at a rate of 30Hz' and they constitute one of the fast-selling consumer products.

Naturally, this spectacular growth is a consequence of its popularity in rapidly growing markets, such as China, India, Africa and Latin America. Historically, GSM was first launched in Finland in 1991 and at the time of writing about 700 mobile networks provide GSM services in more than 200 countries. China, Russia, India and the USA are the countries having the highest number of subscribers. The CEO of the GSM Association, Rob Conway says "What this means is that mobile phones are 'bridging the digital divide' at an astonishing rate with relevant, affordable solutions that help families stay in touch, businesses to grow and economies to develop." This demand has stimulated the design of reliable, robust and yet low-cost GSM handsets.

Given the beneficial economic and social impact of GSM, it is of practical merit to provide a rudimentary introduction to the system's main features for the communications practitioner.

The GSM specifications were released as 13 sets of Recommendations [GSM, 1988], which are summarised in Table 1 covering various aspects of the system [Hanzo, Stefanov, 1999]. The GSM specifications have evolved and are currently managed by the Third-Generation Partnership Project (3GPP). New series numbers (41 to 52) were allocated at Release 4. Each release represents a collection of enhancements to the standard. For example, General Packet Radio Service (GPRS), Enhanced Data Rates for GSM Evolution (EDGE), the GSM/EDGE Radio Access Network (GERAN) and Universal Mobile Telecommunications System (UMTS) were introduced at Release 99 and EDGE Evolution at Release 7. This article considers the originally standardised GSM architecture, which substantially evolved over the years, resulting in a plethora of more sophisticated architectural features that are beyond the scope of this rudimentary overview.

After a brief system overview in Section 1.2 and the introduction of physical and logical channels in Section 1.3 we embark upon describing aspects of mapping logical channels onto physical resources for speech and control channels in Sections 1.4 and 1.5, respectively. These details can be found in Recommendations R.05.02 and R.05.03. Synchronisation issues are considered in Section 1.6. Modulation [R.05.04.], transmission via the standardised wide-band GSM channel models [R.05.05.] as well as adaptive radio link control [R.05.06.], [R.05.08.], discontinuous transmission (DTX) [R.06.31.] and voice activity detection (VAD) [R.06.32.] are to be highlighted in Sections 1.7- 1.10, while a summary of the fundamental GSM features is offered in Section 1.11.

[edit] Overview

The system elements of a GSM public land mobile network (PLMN) are portrayed in Figure 1, where their interconnections via the standardised interfaces A and Um are indicated as well. The mobile station (MS) communicates with the serving and adjacent base stations (BS) via the radio interface Um, while the BSs are connected to the mobile switching centre (MSC) through the network interface A. As seen in Figure 1, the MS includes a Mobile Termination (MT) and a Terminal Equipment (TE). The TE may be constituted for example by a telephone set and fax machine. The MT performs functions needed to support the physical channel between the MS and the base station, such as radio transmissions, radio channel management, channel coding/decoding, speech encoding/decoding, and so forth.

Figure 1: Simplified structure of GSM PLMN © ETT, [Hanzo & Steele, 1994]

The Base Station (BS) is divided functionally into a number of Base Transceiver Stations (BTS) and a Base Station Controller (BSC). The BS is responsible for channel allocation [R.05.09.], link quality and power budget control [R.05.06.], [R.05.08.], signalling and broadcast traffic control, frequency hopping (FH) [R.05.02.], handover (HO) initiation [R.03.09.], [R.05.08.], etc. The MSC represents the gateway to other networks, such as the public switched telephone network (PSTN), integrated services digital network (ISDN) and packet data networks using the interworking functions standardised in [R.09]. The MSC’s further functions include paging, MS location updating [R.03.12.], HO control [R.03.09], etc. The MS’s mobility management is assisted by the home location register (HLR) [R.03.12], storing part of the MS’s location information and routing incoming calls to the visitor location register (VLR) [R.03.12] in charge of the area, where the paged MS roams. Location update is asked for by the MS, whenever it detects from the received and decoded broadcast control channel (BCCH) messages that it entered a new location area. The HLR contains, amongst a number of other parameters, the International Mobile Subscriber Identity (IMSI), which is used for the authentication [R.03.20] of the subscriber by his AUthentication Centre (AUC). This enables the system to confirm that the subscriber is allowed to access it. Every subscriber belongs to a home network and the specific services which the subscriber is allowed to use are entered into his HLR. The Equipment Identity Register (EIR) allows for stolen, fraudulent or faulty mobile stations to be identified by the network operators. The VLR is the functional unit that attends to a MS operating outside the area of its HLR. The visiting MS is automatically registered at the nearest MSC and the VLR is informed of the MSs arrival. A roaming number is then assigned to the MS and this enables calls to be routed to it. The Operations and Maintenance Centre (OMC), Network Management Centre (NMC) and ADministration Centre (ADC) are the functional entities through which the system is monitored, controlled, maintained and managed [R.12].

The MS initiates a call by searching for a BS with a sufficiently high received signal level on the BCCH carrier, it will await and recognise a frequency correction burst and synchronise to it [R.05.08]. Now the BS allocates a bidirectional signalling channel and also sets up a link with the MSC via the network. How the control frame structure assists in this process will be highlighted in Section 1.5. The MSC uses the IMSI received from the MS to interrogate its HLR and sends the data obtained to the serving VLR. After authentication [R.03.20] the MS provides the destination number, the BS allocates a traffic channel and the MSC routes the call to its destination. If the MS moves to another cell, it is re-assigned to another BS and a handover occurs. If both BSs in the handover process are controlled by the same BSC the handover takes place under the control of the BSC, otherwise it is performed by the MSC. In case of incoming calls the MS must be paged by the BSC. A paging signal is transmitted on a paging channel (PCH) monitored continuously by all MSs and covers the location area in which the MS roams. In response to the paging signal the MS performs an access procedure identical to that employed when the MS initiates a call.

[edit] Logical and physical channels

The GSM logical traffic and control channels are standardised in Recommendation [R.05.02], while their mapping onto physical channels is the subject of [R.05.02] and [R.05.03]. The GSM system’s prime objective is to transmit the logical traffic channel’s (TCH) speech or data information. Their transmission via the network requires a variety of logical control channels. The set of logical traffic and control channels defined in the GSM system is summarised in Table 2. There are two general forms of speech and data traffic channels: the full rate traffic channels (TCH/F), which carry information at a gross rate of 22.8 kbit/s, and the half rate traffic channels (TCH/H), which communicate at a gross rate of 11.4 kbit/s. A physical channel carries either a full rate traffic channel, or two half rate traffic channels. In the former the traffic channel occupies one timeslot, while in the latter the two half-rate traffic channels are mapped onto the same timeslot, but in alternate frames.

For a summary of the logical control channels carrying signalling or synchronisation data see Table 2. There are four categories of logical control channels, known as the broadcast control channel (BCCH), the common control channel (CCCH), the stand-alone dedicated control channel (SDCCH) and the associated control channel (ACCH). The purpose and way of deployment of the logical traffic and control channels will be explained by highlighting, how they are mapped onto physical channels in assisting high-integrity communications.

A physical channel in a Time Division Multiple Access (TDMA) system is defined as a timeslot with a timeslot number (TN) in a sequence of TDMA frames. However, the GSM system deploys TDMA combined with frequency hopping (FH) and hence the physical channel is partitioned in both time and frequency. Frequency hopping [R.05.02.] combined with interleaving is known to be very efficient in combating channel fading, and it results in near-Gaussian performance even over hostile Rayleigh-fading channels. The principle of Frequency Hopping (FH) is that each TDMA burst is transmitted via a different RF CHannel (RFCH). If the present TDMA burst happened to be in a deep fade, then the next burst most probably will not be. Consequently the physical channel is defined as a sequence of radio frequency channels and timeslots. Each carrier frequency supports 8 physical channels mapped onto 8 timeslots within a TDMA frame. A given physical channel always uses the same timeslot number TN in every TDMA frame. Therefore, a timeslot sequence is defined by a timeslot number TN and a TDMA frame number (FN) sequence.

Figure 2:

[edit] Speech and data transmission

The speech coding standard is [R.06.10.], while issues of mapping the logical speech traffic channel’s information onto the physical channel constituted by a timeslot of a certain carrier are specified in [R.05.02.]. The original full- and half- rate codecs have been augmented by the introduction of the Enhanced Full Rate (EFR) and Adaptive Multi-Rate (AMR) codecs [1]. Since the error correction coding represents part of this mapping process, also [R.05.03.] is relevant to these discussions. The example of the full rate speech traffic channel (TCH/FS) is here used to highlight, how this logical channel is mapped onto the physical channel constituted by a so-called Normal Burst (NB) of the TDMA frame structure. This mapping is explained by referring to Figure 2 and Figure 3. Then this example will be extended to other physical bursts such as the Frequency Correction- (FCB), Synchronisation- (SB), Access- (AB) and Dummy-Burst (DB) carrying logical control channels, as well as to their TDMA frame structures, as seen in Figure 2 and Figure 6.

The Regular Pulse Excited (RPE) speech encoder is fully characterised in the following References [Vary, Sluyter, 1986], [Hanzo, Somerville, Woodard, 2007] and [Hanzo, Stefanov, 1999]. Due to its complexity its description is beyond the scope of this treatise. Suffice to say that, as it can be seen in Figure 3, it delivers 260 bits/20 ms at a bitrate of 13 kbit/s, which are devided into three significance classes: Class 1a (50 bits), Class 1b (132 bits) and Class 2 (78 bits). The Class 1a bits are encoded by a systematic (53,50) cyclic error detection code by adding three parity bits. Then the bits are reordered and four zero tailing bits are added to perodically reset the memory of the subsequent half rate, constraint length five convolutional codec (CC) CC(2,1,5), as portrayed in Figure 3. Now the unprotected 78 Class 2 bits are concatenated to yield a block of 456 bits/20 ms, which implies an encoded bitrate of 22.8 kbit/s. This frame is partitioned into eight 57-bit subblocks that are blockdiagonally interleaved before undergoing intraburst interleaving. At this stage each 57-bit subblock is combined with a similar subblock of the previous 456-bit frame to construct a 116-bit burst, where the flag bits hl and hu are included to classify, whether the current burst is really a TCH/FS burst or it has been ’stolen’ by an urgent fast associated (FACCH) control channel message. Now the bits are encrypted and positioned in a Normal Burst (NB), as depicted at the bottom of Figure 2, where three tailing bits (TB) are added at both ends of the burst to reset he memory of the Viterbi channel equaliser (VE), which is responsible for removing both the channel-induced and the intentional controlled intersymbol interference [Steele, 1992].

Figure 3: The GSM TDMA frame structure © ETT, [Hanzo & Steele, 1994]

Figure 4: Mapping the TCH/FS logical channel onto a physical channel, © ETT, [Hanzo & Steele, 1994]

The 8.25 bit-interval duration guard period (GP) at the bottom of Figure 2 is provided to prevent burst overlapping due to propagation delay fluctuations. Finally, a 26-bit equaliser training segment is included in the centre of the normal traffic burst. This segment is constructed by a 16-bit Viterbi channel equaliser training pattern surrounded by five quasi-periodically repeated bits on both sides. Since the MS has to be informed about which BS it communicates with, for neighbouring BSs one of eight different training patterns is used, associated with the so-called BS colour codes, which assist in identifying the BSs. This 156.25 bit duration TCH/FS normal burst (NB) constitutes the basic timeslot of the TDMA frame structure, which is input to the Gaussian minimum shift keying (GMSK) modulator to be highlighted in Section 1.7, at a bitrate of approximately 271 kbit/s. Since the bit interval is 1/(271 kbps) = 3.69 \(\mu\)s, the timeslot duration is \(156.25 \cdot 3.69 \approx 0.577\) ms. Eight such normal bursts of eight appropriately staggered TDMA users are multiplexed onto one (RF) carrier giving, a TDMA frame of \(8 \cdot 0.577 \approx 4.615\) ms duration, as shown in Figure 2. The physical channel as characterised above provides a physical timeslot with a throughput of 114 bits/4.615 ms=24.7 kbit/s, which is sufficiently high to transmit the 22.8 kbit/s TCH/FS information. It even has a 'reserved' capacity of 24.7-22.8=1.9 kbit/s, which can be exploited to transmit slow control information associated with this specific traffic channel, i.e., to construct a so-called Slow Associated Control Channel (SACCH), constituted by the SACCH TDMA frames, interspersed with traffic frames at multiframe level of the hierarchy, as seen in Figure 2.

Mapping logical data traffic channels onto a physical channel is essentially carried out by the channel codecs [Wong, Hanzo, 1992], as specified in [R.05.03.]. The full- and half-rate data traffic channels standardised in the GSM system are: TCH/F9.6 , TCH/F4.8, TCH/F2.4, as well as TCH/H4.8, TCH/H2.4, as was portrayed earlier in Table 2. Note that the numbers in these acronyms represent the data transmission rate in kbps. Without considering the details of these mapping processes we now focus our attention on control signal transmission issues.

[edit] Transmission of control signals

The exact derivation, FEC coding and mapping of logical control channel information is beyond the scope of this treatise, the interested reader is referred to [R.05.02.], [R.05.03.] and [Hanzo, Stefanov, 1999] for a detailed discussion. As an example, the mapping of the 184-bit slow associated control channel (SACCH), fast associated control channel (FACCH), broadcast control channel (BCCH), standalone dedicated control channel (SDCCH), paging channel (PCH) and access grant control channel (AGCH) messages onto a 456-bit block, i.e. onto four 114-bit bursts is demonstrated in Figure 4. A double-layer concatenated FIREcode/convolutional code scheme generates 456 bits, using an overall coding-rate of R = 184/456, which gives a stronger protection for control channels than the error protection of traffic channels.

Figure 5: FEC in SACCH, FACCH, BCCH, SDCCH, PCH and AGCH © ETT, [Hanzo & Steele, 1994]

Upon returning to Figure 2 we will now show, how the SACCH is accommodated by the TDMA framestructure. The TCH/FS TDMA frames of the eight users are multiplexed into multiframes of 24 TDMA frames, but the 13\(^{th}\) frame will carry a SACCH message, rather than the 13\(^{th}\) TCH/FS frame, while the 26\(^{th}\) frame will be an idle or dummy frame, as seen at the left hand side of Figure 2 at the multiframe level of the traffic channel hierarchy. The general control channel frame structure shown at the right of Figure 2 is discussed later. This way 24 TCH/FS frames are sent in a 26-frame multiframe during \(26 \cdot 4.615=120\)ms. This reduces the traffic throughput to \(\frac{24}{26} \cdot 24.7=22.8\) kbit/s required by TCH/FS, allocates \(\frac{1}{26} \cdot 24.7=950\)bps to the SACCH and ’wastes’ 950 bps in the idle frame. Observe that the SACCH frame has eight timeslots to transmit the eight 950 bps SACCHs of the eight users on the same carrier. The 950 bps idle capacity will be used in case of half rate channels, where 16 users will be multiplexed onto alternate frames of the TDMA structure to increase system capacity. Then sixteen 11.4 kbit/s encoded half-rate speech TCHs will be transmitted in a 120 ms multiframe, where also sixteen SACCHs are available.

The Fast Associated Control Channel (FACCH) messages are transmitted via the physical channels provided by bits ’stolen’ from their own host traffic channels. The construction of the FACCH bursts from 184 control bits is identical to that of the SACCH, as also shown in Figure 4, but its 456-bit frame is mapped onto eight consecutive 114-bit TDMA traffic bursts, exactly, as specified for TCH/FS. This is carried out by stealing the even bits of the first four and the odd bits of the last four bursts, which is signalled by setting \(hu=1, hl=0\) and \(hu=0, hl=1\) in the first and last bursts, respectively. The unprotected FACCH information rate is 184 bits/20 ms=9.2 kbit/s, which is transmitted after concatenated error protection at a rate of 22.8 kbit/s. The repetition delay is 20 ms and the interleaving delay is \(8 \cdot 4.615=37\)ms, resulting in a total of 57 ms delay.

In Figure 2 at the next hierarchical level 51 TCH/FS multiframes are multiplexed into one superframe lasting \(51 \cdot 120 ms=6.12\) s, which contains \(26 \cdot 51=1326\) TDMA frames. However, in case of 1326 TDMA frames the frame number would be limited to \(0 \leq FN \leq 1326\) and the encryption rule relying on such a limited range of FN values would not be sufficiently secure. Whence 2048 superframes were amalgameted to form a hyperframe of \(1326 \cdot 2048=2 715 648\) TDMA frames lasting \(2048 \cdot 6.12 s \approx 3 h 28 min\ ,\) allowing a sufficiently high FN value to be used in the encryption algorithm. The uplink and downlink traffic-frame structures are identical with a shift of three timeslots between them, which relieves the MS from having to transmit and receive simultaneously, preventing high-level transmitted power leakage back to the sensitive receiver. The received power of adjacent BSs can be monitored during unallocated timeslots.

In contrast to duplex traffic and associated control channels, the simplex BCCH and CCCH logical channels of all MSs roaming in a specific cell share the physical channel provided by timeslot zero of the socalled BCCH carriers available in the cell. Furthermore, as demonstrated by the right hand side section of Figure 2, 51 BCCH and CCCH TDMA frames are mapped onto a \(51 \cdot 4.615=235\) ms duration multiframe, rather than on a 26-frame, 120 ms duration multiframe. In order to compensate for the extended multiframe length of 235 ms, 26 multiframes constitute a 1326-frame superframe of 6.12 s duration. Note in Figure 5 that the allocation of the uplink and downlink frames is different, since these control channels exist only in one direction.

Specifically, the random access channel (RACH) is only used by the MSs in uplink direction if they request, for example, a bidirectional stand-alone dedicated control channel (SDCCH) to be mapped onto an RF channel to register with the network and set up a call. The uplink RACH has a low capacity, carrying messages of eight bits per 235 ms multiframe, which is equivalent to an unprotected control information rate of 34 bps.These messages are concatenated forward error correction (FEC) coded to a rate of 36 bits/235 ms=153 bps. They are not transmitted by the Normal Bursts (NB) derived for TCH/FS, SACCH or FACCH logical channels, but by the so-called Access Bursts (AB), depicted in Figure 6 in comparison to a NB and other types of bursts to be described later. The FEC coded, encrypted 36-bit AB messages of Figure 6, containing amongst other parameters also the encoded 6-bit BS identifier code (BSIC) constituted by the 3-bit PLMN colour code and 3-bit BS colour code for unique BS identification. These 36 bits are positioned after the 41-bit synchronisation sequence, which has a high wordlength in order to ensure reliable access burst recognition and a low probability of being emulated by interfering stray data. These messages have no interleaving delay, while they are transmitted with a repetition delay of one control multiframe length, i.e. 235 ms.

Figure 6: The control multiframe © ETT, [Hanzo & Steele, 1994]

Figure 7: GSM burst structures © ETT, [Hanzo & Steele, 1994]

Adaptive time frame alignment is a technique designed to equalize propagation delay differences between MSs at different distances. The GSM system is designed to allow for cell sizes up to 35 km radius. The time a radio signal takes to travel the 70 km from the base station to the mobile station and back again is \(233.3 \mu s\ .\) As signals from all the mobiles in the cell must reach the base station without overlapping each other, a long guard period of 68.25 bits \((252 \mu s)\) is provided in the access burst, which exceeds the maximum possible propagation delay of \(233.3 \mu s\ .\) This long guard period in the access burst is needed when the mobile station attempts its first access to the base station, or after a handover has occurred. When the base station detects a 41-bit random access synchronisation sequence with a long guard period, it measures the received signal delay relative to the expected signal from a mobile station of zero range. This delay, called the timing advance, is signalled using a 6-bit number to the mobile station, which advances its timebase over the range of 0 to 63 bits, i.e., in units of \(3.69 \mu s\ .\) By this process the TDMA bursts arrive at the BS in their correct timeslots and do not overlap with adjacent ones. This process allows the guard period in all other bursts to be reduced to \(8.25 \cdot 3.69~\mu\)s \(\approx 30.46~\mu\)s (8.25 bits) only. During normal operation the BS continously monitors the signal delay from the MS and if necessary, it will instruct the MS to update its time advance parameter. In very large traffic cells there is an option to actively utilise every second timeslot only to cope with higher propagation delays, which is spectrally inefficient, but in these large, low-traffic rural cells it is admissible.

An example downlink multiframe including a number of BCCH and CCCH logical channels, as transmitted by the BS, is shown in Figure 5. In particular, the last frame is an idle frame (I), while the remaining 50 frames are divided in five blocks of ten frames, where each block starts with a frequency correction channel (FCCH) followed by a synchronisation channel (SCH). In the first block of ten frames the FCH and SCH frames are followed by four broadcast control channel (BCCH) frames and by either four access grant control channels (AGCH) or four paging channels (PCH). In the remaining four blocks of ten frames the last eight frames are devoted to either PCHs or AGCHs, which are mutually exclusive for a specific MS being either paged or granted a control channel.

The frequency correction channel (FCCH), synchronisation channel (SCH) and random access channel (RACH) require special transmission bursts, tailored to their missions, as depicted in Figure 6. The FCCH uses frequency correction bursts (FCB) hosting a specific 142-bit pattern. In partial response GMSK it is possible to design a modulating data sequence, which results in a near-sinusoidal modulated signal immitating an unmodulated carrier exhibiting a fixed frequency offset from the RF carrier utilised. The synchronisation channel transmits synchronisation bursts (SB) hosting a \(16 \cdot 4=64\) bit extended sequence exhibiting a high correlation peak in order to allow frame alignment with a quarter bit accuracy. Furthermore, the SB contains \(2 \cdot 39 = 78\) encrypted FEC-coded synchronisation bits, hosting the BS and PLMN colour codes, each representing one of eight legitimate identifiers. Lastly, the access bursts (AB) contain an extended 41-bit synchronisation sequence and they are invoked to facilitate initial access to the system. Their long guard space of 68.25 bit duration prevents frame overlap, before the MS’s distance, i.e. the propagation delay becomes known to the BS and could be compensated for by adjusting the MS’s timing advance.

[edit] Synchronisation issues

Although some synchronisation issues are standardised in [R.05.02.], [R.05.03.], the GSM Recommendations do not specify the exact BS-MS synchronisation algorithms to be used, these are left to the equipment manufacturers. However, a unique set of timebase counters is defined in order to ensure perfect BS-MS synchronism. The BS sends frequency correction bursts (FCB) and synchronisation bursts (SB) on specific timeslots of the BCCH carrier to the MS to ensure that the MS’s frequency standard is perfectly aligned with that of the BS, as well as to inform the MS about the required initial state of its internal counters. The MS transmits its uniquely numbered traffic and control bursts staggered by three timeslots with respect to those of the BS to prevent simultaneous MS transmission and reception, and also takes into account the required timing advance (TA) to cater for different BS-MS-BS round-trip delays.

The timebase counters used to uniquely describe the internal timing states of BSs and MSs are the Quarter bit Number (QN\(=0 \ldots 624)\) counting the quarter bit intervals in bursts, Bit Number (BN\(=0 \ldots 156)\ ,\) Timeslot Number \((TN=0 \ldots 7)\) and TDMA Frame Number \((FN=0 \ldots 26 \cdot 51 \cdot 2048)\ ,\) given in the order of increasing interval duration. The MS sets up its timebase counters after receiving a SB by determining QN from the 64-bit extended training sequence in the centre of the SB, setting TN=0 and decoding the 78 encrypted, protected bits carrying the 25 SCH control bits.

Figure 8: Synchronisation channel(SCH) message format © ETT, [Hanzo & Steele, 1994]

The SCH carries frame synchronisation information as well as BS identification information to the MS, as seen in Figure 7, and it is provided solely to support the operation of the radio subsystem. The first six bits of the 25-bit segment consist of three PLMN colour code bits and three BS colour code bits supplying a unique BS Identifier Code (BSIC) to inform the MS, which BS it is communicating with. The second 19-bit segment is the so-called Reduced TDMA Frame Number (RFN) derived from the full TDMA Frame Number (FN), constrained to the range of \([0 \ldots (26 \cdot 51 \cdot 2048)-1]=[0 \ldots 2,715,647]\) in terms of three subsegments T1, T2 and T3. These subsegments are computed as follows: T1(11 bits) = [FN div \((26 \cdot 51)\)], T2(5 bits) = [FN mod 26] and T3’(3 bits) = [(T3-1) div 10], where T3 = [FN mod 5], where div and mod represent the integer division and modulo operations, respectively. Explicitly, in Figure 7 T1 determines the superframe index in a hyperframe, T2 the multiframe index in a superframe, T3 the frame index in a multiframe, while T3’ is the so-called signalling block index \([1 \ldots 5]\) of a frame in a specific 51-frame control multiframe, and their role is best understood by referring to Figure 2. Once the MS has received the Synchronisation Burst (SB), it readily computes the FN required in various control algorithms, such as encryption, handover, etc., as shown below\[FN = 51[(T3-T2)\ mod\ 26] + T3 + 51 \cdot 26 \cdot T1, \;\;\; \mbox{where }\;\;\; T3 = 10 \cdot T3'+1.\qquad\qquad(1)\]

[edit] Gaussian minimum shift keying modulation

The GSM system uses constant envelope partial response GMSK modulation [Steele, Hanzo 1999] specified in Recommendation [R.05.04.]. Constant envelope, continuous phase modulation schemes are robust against signal fading as well as interference and have good spectral efficiency. The slower and smoother are the phase changes, the better is the spectral efficiency, since the signal is allowed to change less abruptly, requiring lower frequency components. However, the effect of an input bit is spread over several bit periods, leading to a so-called partial response system, which requires a channel equaliser in order to remove this controlled, intentional intersymbol interference (ISI) even in the absence of uncontrolled channel dispersion.

Figure 9: GMSK modulator schematic diagram © ETT, [Hanzo & Steele, 1994]

The widely employed partial response GMSK scheme is derived from the full response Minimum Shift Keying (MSK) scheme. In MSK the phase changes between adjacent bit periods are piecewise linear, which results in discontinuous phase derivative, i.e., instantaneous frequency at the signalling instants, and hence widens the spectrum. However, smoothing these phase changes by a filter having a Gaussian impulse response [Steele, 1992], which is known to have the lowest possible bandwidth, this problem is circumvented using the schematic of Figure 8, where the GMSK signal is generated by modulating and adding two quadrature carriers. The key parameter of GMSK in controlling both bandwidth and interference resistance is the 3 dB-down filter-bandwidth \(\times\) bit interval product \((B \cdot T)\) referred to as normalised bandwidth. It was found that as the \(B \cdot T\) product is increased from 0.2 to 0.5, the interference resistance is improved by approximately 2 dB at the cost of increased bandwidth occupancy, and best compromise was achieved for \(B \cdot T=0.3\ .\) This corresponds to spreading the effect of one bit over approximately three bit intervals. The spectral efficiency gain due to higher interference tolerance and hence more dense frequency reuse was found to be more significant than the spectral loss caused by wider GMSK spectral lobes. Finally, it is worth noting that during the initial call-setup phase no coherent detection is feasible, hence differential encoding and differentially coherent detection are employed.

The channel separation at the TDMA burst-rate of 271kbit/s is 200kHz and the modulated spectrum must be 40dB down at both adjacent carrier frequencies. When TDMA bursts are transmitted in an on-off keyed mode, further spectral spillage arises, which is mitigated by a smooth power ramp-up and down envelope at the leading and trailing edges of the transmission bursts, attenuating the signal by 70dB during a 28 \(\mu\)s and 18 \(\mu\)s interval, respectively.

[edit] Wideband channel models

Figure 10: Typical GSM channel impulse responses © ETT, [Hanzo & Steele, 1994]

The set of 6-tap GSM impulse responses [Greenwood, Hanzo, 1999] specified in Recommendation [R.05.05.] is depicted in Figure 9, where the individual propagation paths are independent Rayleigh fading paths, weighted by the appropriate coefficients \(h_i\) corresponding to their relative powers portrayed in the Figure. In simple terms the wideband channel’s impulse response is measured by transmitting an impulse and detecting the received echos at the channel’s output in every D-spaced so-called delay bin. In some bins no delayed and attenuated multipath component is received, while in others significant energy is detected, depending on the typical reflecting objects and their distance from the receiver. The path-delay can be easily related to the distance of the reflecting objects, since radio waves are travelling at the speed of light. For example, at a speed of 300 000 km/s a reflecting object situated at a distance of 0.15 km yields a multipath component at a round-trip delay of 1 \(\mu\)s.

The Typical Urban (TU) impulse response spreads over a delay interval of 5 \(\mu\)s, which is almost two 3.69 \(\mu\)s bit intervals duration and therefore results in serious InterSymbol Interference (ISI). Whence in simple terms it can be treated as a two-path model, where the reflected path has a length of 0.75 km, corresponding to a reflector located at a distance of about 375 m. The Hilly Terrain (HT) model has a sharply decaying shortdelay section due to local reflections and a long-delay path around 15 \(\mu\)s due to distant reflections. Therefore in practical terms it can be considered a two- or three-path model having reflections from a distance of about 2 km. The Rural Area (RA) response seems the least hostile amongst all standardised responses, decaying rapidly inside one bit interval and therefore is expected to be easily combated by the channel equaliser. Although the type of the equaliser is not standardised, partial response systems typically use Viterbi Equalisers (VE). Since the RA channel effectively behaves as a single-path non-dispersive channel, it would not require an equaliser. The fourth standardised impulse response is artificially contrived in order to test the equaliser’s performance and is constituted by six equidistant unit-amplitude impulses representing six equal-powered independent Rayleighfading paths with a delay-spread over 16 \(\mu\)s. With these impulse responses in mind the required channel is simulated by summing the appropriately delayed and weighted received signal components. In all but one cases the individual components are assumed to have Rayleigh amplitude distribution, while in the RA model the main tap at zero delay is supposed to have Rician distribution with the presence of a dominant line-of-sight path.

Figure 11: Initial cell selection by the MS © ETT, [Hanzo & Steele, 1994]

[edit] Adaptive link control

The adaptive link control algorithm portrayed in Figure 10 and specified in [R.05.08.] allows for the MS to favour that specific traffic cell, which provides the highest probability of reliable communications associated with the lowest possible path loss. It also decreases interference with other co-channel users and, through dense frequency reuse, improves spectral efficiency, whilst maintaining an adequate communications quality and facilitates a reduction in power consumption, which is particularly important in hand-held MS’s. The handover process maintains a call in progress as the MS moves between cells, or when there is an unacceptable transmission quality degradation caused by interference, in which case an intra-cell handover to another carrier in the same cell is performed. A radio link failure occurs when a call with an unacceptable voice or data quality cannot be improved either by RF power control or by handover. The reasons for the link failure may be loss of radio coverage or very high interference levels. The link control procedures rely on measurements of the received RF signal strength (RXLEV), the received signal quality (RXQUAL), and the absolute distance between base and mobile stations (DISTANCE).

RXLEV is evaluated by measuring the received level of the broadcast control channel (BCCH) carrier, which is continuously transmitted by the BS on all time slots of the B frames in Figure 5 and without variations of the RF level. A MS measures the received signal level from the serving cell and from the BS’s in all adjacent cells by tuning and listening to their BCCH carriers. The root mean squared (RMS) level of the received signal is measured over a dynamic range of -103 to -41 dBm for intervals of one SACCH multiframe (480 ms). The received signal level is averaged over at least 32 SACCH frames \((\approx 15 s)\) and mapped to give RXLEV values between 0 and 63 to cover the range \(-103 \ldots -41\) dBm in steps of 1 dB. The RXLEV parameters are then coded into 6-bit words for transmission to the serving BS via the SACCH.

RXQUAL is estimated by measuring the BER before channel decoding, using the Viterbi channel equaliser's metrics [Steele, 1992] and/or those of the Viterbi convolutional decoder [Wong, Hanzo, 1992]. Eight values of RXQUAL span the logarithmically scaled BER range of \(0.2% \ldots 12.8%\) before channel decoding.

The absolute DISTANCE between base and mobile stations is measured using the "timing advance" parameter. The timing advance is coded as a 6 bit number corresponding to a propagation delay from 0 to \(63 \cdot 3.69 \mu s=232.6 \mu\)s, characteristic of a cell radius of 35km.

While roaming, the MS needs to identify which potential target BS it is measuring and the BCCH carrier frequency may not be sufficient for this purpose, since in small cluster sizes the same BCCH frequency may be used in more than one surrounding cell. To avoid ambiguity a 6-bit Base Station Identity Code (BSIC) is transmitted on each BCCH carrier in the synchronisation burst (SB) of Figure 6. Two other parameters transmitted in the BCCH data provide additional information about the BS. The binary flag called PLMN PERMITTED indicates whether the measured BCCH carrier belongs to a PLMN which the MS is permitted to access. The second Boolean flag, CELL BAR ACCESS, indicates whether the cell is barred for access by the MS, although it belongs to a permitted PLMN. A MS in idle mode, i.e., after it has just been switched-on, or after it has lost contact with the network, searches all 125 RF channels and takes readings of RXLEV on each of them. Then it tunes to the carrier with the highest RXLEV and searches for frequency correction bursts (FCB) in order to determine whether or not the carrier is a BCCH carrier. If it is not, then the MS tunes to the next highest carrier, and so on, until it finds a BCCH carrier, synchronises to it and decodes the parameters BSIC, PLMN PERMITTED and CELL BAR ACCESS in order to decide whether to continue the search. The MS may store the BCCH carrier frequencies used in the network accessed, in which case the search time would be reduced. Again, the process described is summarised in the flowchart of Figure 10.

The adaptive power control is based on RXLEV measurements. In every SACCH multiframe the BS compares the RXLEV readings reported by the MS, or obtained by the base station, with a set of thresholds. The exact strategy for RF power control is determined by the network operator with the aim of providing an adequate quality of service for speech and data transmissions while keeping interferences low. Clearly, 'adequate' quality must be achieved at the lowest possible transmitted power to keep cochannel interferences low, which implies contradictory requirements in terms of transmitted power. The criteria for reporting radio link failure are based on the measurements of RXLEV and RXQUAL performed by both the mobile and base stations and the procedures for handling link failures result in the re-establishment or the release of the call, depending on the network operator's strategy.

The handover process involves the most complex set of procedures in the radio link control. Handover decisions are based on results of measurements performed both by the base and mobile stations. The base station measures RXLEV, RXQUAL, DISTANCE, and also the interference level in unallocated time slots, while the MS measures and reports to the BS the values of RXLEV and RXQUAL for the serving cell, and RXLEV for the adjacent cells. When the MS moves away from the BS, the RXLEV and RXQUAL parameters for the serving station become lower, while RXLEV for one of the adjacent cells increases.

[edit] Discontinuous transmission

Discontinuous transmission (DTX) issues are standardised in Recommendation [R.06.31], while the associated problems of voice activity detection (VAD) are specified by [R.06.32.]. Assuming an average speech activity of 50% and a high number of interferers combined with frequency hopping to randomise the interference load, significant spectral efficiency gains can be achieved, when deploying discontinuous transmissions due to decreasing interferences, while reducing power dissipation as well. Due to the reduction in power consumption full DTX operation is mandatory for MSs, but in BSs only receiver DTX functions are compulsory.

The fundamental problem in voice activity detection is how to differentiate between speech and noise, while keeping false noise triggering and speech spurt clipping as low as possible. In vehicle-mounted MSs the severity of the speech/noise recognition problem is aggravated by the excessive vehicle background noise. This problem is resolved by deploying a combination of threshold comparisons and spectral domain techniques [GSM, 1988], [Hanzo, Stefanov, 1992]. Another important associated problem is the introduction of noiseless inactive segments, which is mitigated by comfort noise insertion (CNI) in these segments at the receiver.

[edit] Summary

Following the standardisation and launch of the GSM system its salient features were summarised in this brief review. Time Division Multiple Access (TDMA) with eight users per carrier is used at a multi-user rate of 271 kbit/s, demanding a channel equaliser to combat dispersion in large cell environments. The error protected chip-rate of the full-rate traffic channels is 22.8kbit/s, while in half-rate channels is 11.4kbit/s. Apart from the full- and half-rate speech traffic channels there are five different-rate data traffic channels and 14 various control and signalling channels to support the system’s operation. A moderately complex, 13 kbit/s Regular Pulse Excited speech codec with long term predictor (LTP) is used, combined with an embedded three-class error correction codec and multi-layer interleaving to provide sensitivity-matched unequal error protection for the speech bits. An overall speech delay of 57.5ms is maintained. Slow frequency hopping at 217 hops/sec yields substantial performance gains for slowly moving pedestrians.

Constant envelope partial response GMSK with a channel spacing of 200kHz is deployed to support 124 duplex channels in the primary (P-GSM) 890 − 915MHz up-link and 935 − 960MHz down-link bands, respectively. GSM is deployed worldwide and as a consequence the frequency bands supported by the standard have extended over time. The DCS-1800 (1710.0 – 1785.0MHz up-link 1805.0 – 1880.0MHz down-link) and PCS-1900 (1850.0 – 1910.0MHz up-link 1930.0–1990.0MHz down-link) and the extended (E-GSM) (880.0–915.0MHz up-link 925.0–960.0MHz down-link) predate Release 99. Other bands have since been incorporated into the specifications; for example, GSM-450 (Release 99) and GSM-700 (Release 4). At a transmission rate of 271 kbit/s a spectral efficiency of 1.35 bit/s/Hz is achieved. The controlled GMSK-induced and un-controlled channel-induced inter-symbol interferences are removed by the channel equaliser. The set of standardised wide-band GSM channels was introduced in order to provide bench-markers for performance comparisons. Efficient power budgeting and minimum co-channel interferences are ensured by the combination of adaptive power- and handover-control based on weighted averaging of up to eight uplink and downlink system parameters. Discontinuous transmissions assisted by reliable spectral-domain voice activity detection and comfort-noise insertion further reduce interferences and power consumption. Due to ciphering, no unprotected information is sent via the radio link. As a result, spectrally efficient, high-quality mobile communications with a variety of services and international roaming is possible in cells of up to 35km radius for signal to noise- and interference-ratios in excess of 10 − 12dBs. The key system features are summarised in Table 3.

Table 3: Summary of GSM features

Figure 12:

[edit] References

1. GSM Recommendation, 1988, (Group Speciale Mobile or Global System of Mobile Communication) European Telecommunications Standardisation Institue.
2. R. Steele, L. Hanzo (Ed): Mobile Radio Communications: Second and Third Generation Cellular and WATM Systems, John Wiley-IEEE Press, 2nd edition, July 1999, ISBN 0-471-97806-x, 1060 pages (online)
3. Hanzo, L. and Steele, R., 1994, ”The Pan-European Mobile Radio System, Part 1 and 2”, European Transactions on Telecommunications, Vol. 5, No. 2, Mar.-Apr. 1994, pp 245-276.
4. Vary, P. and Sluyter, R.J., 1986, ”MATS-D Speech Codec: Regular-Pulse Excitation LPC”, Proc. of Nordic Conference on Mobile Radio Communications, pp 257-261.
5. L. Hanzo, F.C.A. Somerville, J.P. Woodard: Voice and Audio Compression for Wireless Communications, John Wiley and IEEE Press, 2007 (online)
6. Steele, R. (Ed.), 1992, Digital mobile radio communication, IEEE Press-Pentech Press, ISBN 07273-1406-8.
7. Greenwood, D. and Hanzo, L., 1992, Characterisation of Mobile Radio Channels , Chapter 2, pp 92-185, in R. Steele (Ed.) Mobile Radio Communications, IEEE Press-Pentech Press, London, ISBN 07273-1406-8
8. Wong, K.H.H. and Hanzo, L., 1992, Channel Coding, Chapter 4, pp 347-488, in R. Steele (Ed.) Mobile Radio Communications, IEEE Press-Pentech Press, London, ISBN 07273-1406-8.
9. L. Hanzo, T.H. Liew, B.L. Yeap: Turbo Coding, Turbo Equalisation and Space-Time Coding, John Wiley, August 2002, ISBN 0-470-84726-3, 766 pages (online)

Internal references

• Jan A. Sanders (2006) Averaging. Scholarpedia, 1(11):1760.

• Olaf Sporns (2007) Complexity. Scholarpedia, 2(10):1623.

• Giovanni Gallavotti (2008) Fluctuations. Scholarpedia, 3(6):5893.

[edit] See also