Advanced telecommunication systems Part I : mobile network dimensioning Salah Eddine El Ayoubi
October 2010
outline
objective: ensuring QoS in mobile networks
dimensioning for ensuring coverage
dimensioning for ensuring capacity – GSM – UMTS – LTE – just before LTE: HSDPA – after LTE: LTE-A
2
Salah Eddine EL AYOUBI – October 2010
outline
objective: ensuring QoS in mobile networks
dimensioning for ensuring coverage
dimensioning for ensuring capacity – GSM – UMTS – LTE – just before LTE: HSDPA – after LTE: LTE-A
3
Salah Eddine EL AYOUBI – October 2010
coverage targets
mobile operators have to ensure complete coverage: – minimize white zones – cover villages as well as cities – cover routes
limited coverage of any base station: – limited power – loss due to propagation
4
Salah Eddine EL AYOUBI – October 2010
cellular networks
each base station covers a cell / sector
large cells required to reduce costs, however: – degraded QoS at cell edge: coverage problems – many users served: capacity problems
5
Salah Eddine EL AYOUBI – October 2010
QoS targets
coverage is not the only criterion: – QoS in coverage areas is important
QoS includes: – access rate – good communication probability – throughput
operator target: – ensure coverage target and QoS – with lowest costs
operator dilemma: – low cost -> large cells -> more users in each cell -> more spectrum needed – spectrum is limited and too costly
6
Salah Eddine EL AYOUBI – October 2010
What is spectrum ? Radio waves are characterized by their frequency, measured in Hertz (Hz)
f1
f2
300 MHz
30 MHz
VHF
f3
3 GHz
UHF
30 GHz
SHF
Spectrum is the continuous aggregation of these frequencies
7
Salah Eddine EL AYOUBI – October 2010
Main guidelines when managing spectrum
Spectrum shall be usable (not all frequencies are valuable for every type of radio access) coverage
Coverage frequencies
Spectrum shall be managed as efficiently as possible
Salah Eddine EL AYOUBI – October 2010
Coverage too small
Terminal too big
400
8
Capacity frequencies
1000
5000
Frequency (MHz)
How it works ?
f3
f3
f1
f2
9
Salah Eddine EL AYOUBI – October 2010
f1
f2
f3
f1
f2
High demand
Limited resource
10
Salah Eddine EL AYOUBI – October 2010
outline
objective: ensuring QoS in mobile networks
dimensioning for ensuring coverage
dimensioning for ensuring capacity – GSM – UMTS – LTE – just before LTE: HSDPA – after LTE: LTE-A
11
Salah Eddine EL AYOUBI – October 2010
link budget
link budget objective maximum distance between a user and its serving base station while guaranteeing a given quality of service
equipment parameters
12
propagation model
Salah Eddine EL AYOUBI – October 2010
received signals
SINR
cell range
equipment parameters
determine gains and losses due to equipments.
antenna gain GA: – directivity of antenna amplifies the signal in some directions.
feeder loss LC: – due to the cable between amplifier and antenna.
body loss LB: – due to the body of the user.
for an emitted power Pmax:
Pmax × G A useful power = LF LB
13
Salah Eddine EL AYOUBI – October 2010
propagation model
link budget objective maximum distance between a user and its serving base station while guaranteeing a given quality of service
equipment parameters
14
propagation model
Salah Eddine EL AYOUBI – October 2010
received signals
SINR
cell range
radio channel channel variations are due to – pathloss attenuation – shadowing (slow fading) – fast fading
Attenuation (dB)
Path Loss Shadowing Fast fading
Distance
15
path loss is due to the distance between the transmitter and the receiver
shadowing is due to the obstacles between the transmitter and the receiver
fast fading is due to multipath propagation (reflections on obstacles that create multiple paths of the received signal)
for coverage dimensioning, focus is on the path loss, adding a margin for shadowing
Salah Eddine EL AYOUBI – October 2010
use of propagation models
Ptx
Ptx pathloss
pathloss C
I
Serving BS
Interfering BS
propagation models allow to compute: – The received signal power (⇒ coverage maps) – The interfering power (⇒ QoS maps)
16
a propagation model is the first building block of (almost) any radio planning tool
Salah Eddine EL AYOUBI – October 2010
path loss models
free space propagation 4πD 4πDf Pathloss = = λ c 2
2
D
– only valid for line of sight, without multimulti-path – these conditions are not met in cellular networks
statistical models (e.g. Okumura-Hata)
Pathloss[dB ] = A + B ⋅ log(D ) with 20 ≤ B ≤ 40
– simple models with A & B statistically tuned for typical environments (urban, etc.) – no geographical data required – useful for dimensioning 17
Salah Eddine EL AYOUBI – October 2010
e.g. urban environment
D
received signals
link budget objective maximum distance between a user and its serving base station while guaranteeing a given quality of service
equipment parameters
18
propagation model
Salah Eddine EL AYOUBI – October 2010
received signals
SINR
cell range
received signals
for a user situated at distance d from a base station: ξ
Pmax × G A 10 10 received power = × LF LB PL(d ) – PL(d)=path loss at distance d – ξ shadowing variable
19
Salah Eddine EL AYOUBI – October 2010
SINR
link budget objective maximum distance between a user and its serving base station while guaranteeing a given quality of service
equipment parameters
20
propagation model
Salah Eddine EL AYOUBI – October 2010
received signals
SINR
cell range
interference in the dowlink
interference is received by the mobile from the base stations: – it depends on the position of the mobile in the cell – cell-edge users are subject to higher interference because they are closer to interferers.
observations: – the origin of interference is well defined. – the intensity of this interference is to be calculated.
21
Salah Eddine EL AYOUBI – October 2010
interference in the uplink
interference is received by the base station from the mobiles in adjacent cells: – it is independent from the position of the mobile in the cell. – it depends on the distribution of mobiles in interfering cells.
observations: – the average interference is uniform for all mobiles. – the position of interferers is unknown.
22
Salah Eddine EL AYOUBI – October 2010
SINR calculations
collisions decrease the Signal to Interference Ratio (SINR):
received power SINR = received interference + noise
a lower SINR means a larger Bit Error Rate (BER): – degraded QoS
23
Salah Eddine EL AYOUBI – October 2010
cell range
link budget objective maximum distance between a user and its serving base station while guaranteeing a given quality of service
equipment parameters
24
propagation model
Salah Eddine EL AYOUBI – October 2010
received signals
SINR
cell range
maximal cell range
for a good reception, the SINR must be larger than a target: – SINR>SINRtarget
for a given cell range R, calculate the SINR at cell edge: – SINR(R) – for a larger R, SINR degrades as received power becomes lower compared to noise
the optimal cell range is the largest R so that – SINR(R)>SINRtarget
25
in general, the limiting link for coverage is the uplink as mobiles have low emitted powers.
Salah Eddine EL AYOUBI – October 2010
example coverage of a cell
exercise
26
Salah Eddine EL AYOUBI – October 2010
outline
objective: ensuring QoS in mobile networks
dimensioning for ensuring coverage
dimensioning for ensuring capacity – GSM – UMTS – LTE – just before LTE: HSDPA – after LTE: LTE-A
27
Salah Eddine EL AYOUBI – October 2010
Erlang-like capacity
need to install resources: – until a target Quality of Service (QoS) is achieved for users – example: number of frequency carriers per cell
user perceived QoS includes: – blocking rates for real-time calls – download time for FTP-like users
this is called Erlang-like capacity: – reference to mathematician Agner Krarup Erlang
– example Erlang-B law.
28
Salah Eddine EL AYOUBI – October 2010
Erlang-B law Erlang table
probability of call loss:
– – – –
B=blocking rate E=traffic intensity C= number of circuits Each call uses one circuit
N
0.0001 0.001
100 95 90 85 80 75 70 65 60 55 50 45 40 35 30 25 20 15 10 5
5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85
A simple Erlang calculator can be found at:
http://perso.rd.francetelecom.fr/bonald/Applets/erlang.html
29
Salah Eddine EL AYOUBI – October 2010
0.01
the race for bit rates in mobile networks Mobility
2000
1995
WIDE AREA MOBILITY
2010
2005
HSDPA HSDPA GSM
EDGE EDGE
GPRS
UMTS UMTS HSUPA
HSPA HSPA
LTE
++
4G? 4G?
Mobile DVB-xTV 802.16m
SHORT RANGE
B3G
MOBILITY
Fixed FixWimax FIXED
WLAN WLAN
Data Rate 10kbps
30
100kbps
Salah Eddine EL AYOUBI – October 2010
1Mbps
10Mbps
100Mbps
outline
objective: ensuring QoS in mobile networks
dimensioning for ensuring coverage
dimensioning for ensuring capacity – GSM – UMTS – LTE – just before LTE: HSDPA – after LTE: LTE-A
31
Salah Eddine EL AYOUBI – October 2010
GSM operation
the spectrum assigned to GSM is divided into sub-bands of 200 KHZ each.
the subbands cannot be used in adjacent cells – due to inter-cell interference – a frequency reuse map is necessary
1/3 of sub-bands used in each cell
32
1/7 of sub-bands used in each cell a transmitter (a dedicated amplifier) is necessary for each subband in the cell.
Salah Eddine EL AYOUBI – October 2010
Time Division Multiple Access operation
several frequency sub-bands of 200 KHZ each
each sub-band is allocated for different users at different times
the time frame of 4.62 ms is divided into 8 time slots – but the transmitter serves up to 7 users (one TS for signalling)
Transmitters
33
Salah Eddine EL AYOUBI – October 2010
Time slots
example capacity of a GSM cell
exercise
34
Salah Eddine EL AYOUBI – October 2010
outline
objective: ensuring QoS in mobile networks
dimensioning for ensuring coverage
dimensioning for ensuring capacity – GSM – UMTS – LTE – just before LTE: HSDPA – after LTE: LTE-A
35
Salah Eddine EL AYOUBI – October 2010
outline: UMTS
36
physical layer
admission control
capacity calculations
Salah Eddine EL AYOUBI – October 2010
Code Division Multiple Access
37
everybody transmits at the same time-frequency resources.
each transmitter has its own code
the receiver decodes the signal and views the others' signals as residual interference.
Salah Eddine EL AYOUBI – October 2010
spreading process
38
Salah Eddine EL AYOUBI – October 2010
downlink spreading codes Walsh code: W(0,1) = 1 W(0,2) = 1, 1 W(1,2) = 1,-1 W(0,4) = 1, 1, 1, 1 W(1,4) = 1,-1, 1,-1 W(2,4) = 1, 1,-1,-1 W(3,4) = 1,-1,-1, 1 W(0,8) = 1, 1, 1, 1, 1, 1, 1, 1 W(1,8) = 1,-1, 1,-1, 1,-1, 1,-1 W(2,8) = 1, 1,-1,-1, 1, 1,-1,-1 W(3,8) = 1,-1,-1, 1, 1,-1,-1, 1 W(4,8) = 1, 1, 1, 1,-1,-1,-1,-1 W(5,8) = 1,-1, 1,-1,-1, 1,-1, 1 W(6,8) = 1, 1,-1,-1,-1,-1, 1, 1 W(7,8) = 1,-1,-1, 1,-1, 1, 1,-1
39
orthogonal codes, as synchronous transmissions
problem: multipath propagation that introduces delays
Salah Eddine EL AYOUBI – October 2010
uplink spreading codes (1/2)
ak
Maximum Length (ML) sequence
sequence determined by the XOR feedbacks.
if register of length R, sequence of period L=2R-1
XOR of a sequence with a shifted version of it gives another version of the same ML sequence.
characterized by irreductible polynom: f ( x) =
c ( n) =
1≤ k ≤ R
∑ a c(n − k ),
c(n + j ) =
k
mod 2
c ( n) ⊕ c ( n + j ) =
∑ a c(n − k + j ) k
mod 2
∑ a c(n − k ) ⊕ c( n − k + j ) k
mod 2
d ( n ) = c ( n) ⊕ c ( n + j ) =
1≤ k ≤ R
∑ a d (n − k ) k
mod 2 40
Salah Eddine EL AYOUBI – October 2010
∑a x k
mod 2
1≤ k ≤ R
1≤ k ≤ R
0≤ k ≤ R
k
uplink spreading codes (2/2)
inter-correlation between ML sequences may be large.
for obtaining good correlation properties, Gold codes are generated by EXOR-ing some preferred pairs of ML-sequences
Gold demonstrates that, if we choose carefully two ML sequences of length L=2R-1, characterized by polynoms f(x) and g(x), such that inter-correlation is low, the ML sequences of length L generated by z(x)=f(x).g(x) have also low correlation.
not orthogonal but with low correlation for cases where transmitters are not synchronized
R=6
f(x)=x6+x+1, g(x)=x6+x5+x2+x+1
z(x)=x12+x11+x8+2x7+3x6+x5+x3+2x2+2x+1 =x12+x11+x8+x6+x5+x3+1
sequence of 22R-1, divided into 2R+1 sequences of length L. 41
Salah Eddine EL AYOUBI – October 2010
dealing with inter-cell interference
42
scrambling codes (Gold code) separate also cells in the downlink.
inter-cell interference is reduced as if it were a transmission from the same cell.
Salah Eddine EL AYOUBI – October 2010
outline: UMTS
43
physical layer
admission control
capacity calculations
Salah Eddine EL AYOUBI – October 2010
downlink SINR model 1.
r0
Pmax to share
44
Salah Eddine EL AYOUBI – October 2010
the SINR for a mobile depends on the distance r0 from the BS, as inter-cell interference increases at cell edge.
cell decomposition
45
Salah Eddine EL AYOUBI – October 2010
1.
the SINR for a mobile depends on the distance r0 from the BS, as inter-cell interference increases at cell edge.
2.
to simplify the problem, divide the cell into concentric rings
3.
a mobile is thus charcterized by its service and its position in the cell.
4.
calculate powers and SINRs.
5.
apply admission control: emitted power< maximal power.
emitted power
zone i is characterized by: – path loss qi,l with cell l – interference factor Fi =
qi ,0
∑q l ≠0
i ,l
service c characterized by target quality: β c = – S: spreading factor
SINR c S + α .SINR c
multi-path propagation introduces a orthogonality factor α
a power PCom is used for signalling
adjacent cells have average load χ
number of users of class c in zone i is Mi,c
the total transmitted power is
PCom +
Ptot =
n
C
∑ (χ P
max Fi
i =1
1−α
Salah Eddine EL AYOUBI – October 2010
∑β M c
c =1
n
C
∑ (∑ β M c
i =1
46
+ N 0 qi )(
c =1
i ,c )
i ,c )
admission control
power of base station limited by Pmax
admission control constraint: n
∑ (αP
max
+ χ Pmax Fi + N 0 qi )(
i =1
C
∑β M c
c =1
intra-cell interference noise intra-cell interference
47
Salah Eddine EL AYOUBI – October 2010
i ,c )
≤ Pmax − PCom
outline: UMTS
48
physical layer
admission control
capacity calculations
Salah Eddine EL AYOUBI – October 2010
capacity calculations
admission control constraint indicates that there is a resource (power) shared by users of different demands (position+service).
traffic ρc,i (Erlang) in zone i for class c.
multi-Erlang analysis is suitable:
1 Pr[ M 1,1 ,..., M C ,n ] = G
49
Salah Eddine EL AYOUBI – October 2010
C
n
∏∏ c =1 i =1
ρ c,i M c ,i M c ,i !
capacity calculations
Exercise
50
Salah Eddine EL AYOUBI – October 2010
outline
objective: ensuring QoS in mobile networks
dimensioning for ensuring coverage
dimensioning for ensuring capacity – GSM – UMTS – LTE – just before LTE: HSDPA – after LTE: LTE-A
51
Salah Eddine EL AYOUBI – October 2010
outline: LTE
52
physical layer
throughput calculations
capacity calculations
use case: mobile TV
Salah Eddine EL AYOUBI – October 2010
Beyond 3G context and E-UTRAN requirements Expected performance (based on analysis and simulations) Peak rate (Downlink) (in 20 MHz, FDD)
144 Mbit/s
2 Tx and 2 Rx antennas, 64 QAM modulation, code rate 5/6
56 Mbit/s (71 Mbit/s for 64QAM)
1 Tx antenna, 2 Rx antennas 16 QAM modulation, code rate 5/6
Average cell spectrum efficiency (downlink)
1.72 b/s/Hz/cell (8.6 Mbit/s in 5 MHz)
2 Tx and 2 Rx antennas MIMO transmission with linear receiver
Average cell spectrum efficiency (uplink)
0.7 b/s/Hz/cell (3.5 Mbit/s in 5 MHz)
2 Tx and 2 Rx antennas No Multi-user - MIMO
User plane latency (two way radio delay)
~ 10 ms
Assumptions: FDD, 30% retransmissions
Peak rate (Uplink) (in 20 MHz, FDD)
Connection setup latency
53
< 50 msecs (dormant->active) < 100 msecs (idle ->active)
Salah Eddine EL AYOUBI – October 2010
the 3M of Beyond 3G similar principles are used by most beyond 3G air interfaces - the physics are the same for everybody !
54
Multi-carrier
Multi-antenna (MIMO)
Multi-Layer
– Frequency dimension – Allow for spectrum flexibility and higher bandwidths. – Data rate = Bandwidth [Hz] x Spectrum efficiency [bps/Hz]
– Spatial dimension – Higher spectrum efficiencies – Information Theory: Max. spectrum efficiency increases linearly with the number of antennas.
– Cross-layer optimization (PHY, MAC, RLC…) – Packet oriented radio interface – Low latencies and higher spectrum efficiencies.
Salah Eddine EL AYOUBI – October 2010
fast fading parameters (1/3)
fundamental parameters of the fast fading channel Remote Scatterer Local-to-mobile Scatterers
- delay spread (frequency selectivity) - maximum delay: tmax - coherence band: Bc = 1/tmax
Terminal v Basestation
- Bc=maximum bandwidth over which two frequencies of a signal are likely to experience correlated fast fading.
Remote Scatterer
- if the symbol duration is much larger than tmax, impact of delay spread is negligible.
tmax 55
Salah Eddine EL AYOUBI – October 2010
fast fading parameters (2/3)
fundamental parameters of the fast fading channel -Doppler spread (time selectivity)
Remote Scatterer Local-to-mobile Scatterers
- Mobile speed v - serving frequency fC
Terminal v Basestation Remote Scatterer
- Maximum doppler: fD = fC x v/c0 - Coherence time: Tc = 1 / (2 fD)
- signal arrives at the receiver within the interval [fC-fD,fC+fD] - if the baseband signal bandwidth is much greater than fD the effects of Doppler spread are negligible.
56
Salah Eddine EL AYOUBI – October 2010
fast fading parameters (3/3)
fundamental parameters of the fast fading channel - angle spread (spatial selectivity)
Remote Scatterer Local-to-mobile Scatterers
Terminal v Base station Remote Scatterer
- difference in angles of arrival/departure - coherence distance is the maximum spatial separation over which the channel response can be assumed constant.
-for small angle spread, coherence distance is large -for large angle spread, coherence distance is small (e.g. in mobile communications).
57
Salah Eddine EL AYOUBI – October 2010
multi-carrier … the frequency dimension
Orthogonal Frequency Division Multiplexing (OFDM) – Facilitates equalization at the receiver – Divides bandwidth in narrowband sub-carriers – Simple frequency domain equalization
– Time-frequency resources can be allocated to data and control channels – Various spectrum allocations can be addressed with the same technology
– E-UTRAN uses Single –Carrier FDMA (SC-FDMA)
58
Salah Eddine EL AYOUBI – October 2010
L1/L2 Control
User A
User B
Spectrum allocation 1.25 - 20 MHz
– Modified scheme may be needed in uplink – Similar properties than OFDM, but allows for cheap power amplifiers at the terminal.
Frequency
– OFDM Access (OFDMA) provides flexibility for resource allocation
Time
1ms sub-frame (LTE DL)
h*0
User K Modulation Coding
+ TG
- TG
FFT
Symbol mapping
. . .
Coding
S/P
0
P/S
User 1 Modulation
IFFT
multi-carrier … the frequency dimension OFDM parameters and signal design
h*Nc-1
Symbol demapping
NC -1
Nc narrowband sub-carriers
design rules – Avoid inter symbol interference: Guard interval (TG) > Maximum Channel delay (tmax) – Avoid inter carrier interference: Carrier spacing (∆f=1/TS) >> max. Doppler spread (2fD) – Limit overhead and ensure time invariance: TG ~0.25TS, TS+TG << Tc – Number of carriers is around 60-80% of FFT size to ensure spectrum emission mask.
59
Salah Eddine EL AYOUBI – October 2010
Multi-carrier … the frequency dimension Basic parameters of E – UTRAN Downlink
Frequency
L1/L2 Control
User A
User B
Time
20 MHz
1ms sub-frame (LTE DL)
60
Salah Eddine EL AYOUBI – October 2010
Multi-carrier … the frequency dimension
SC - FDMA signal design g*0
0 + TG
0 NC -1
SC - FDMA properties – Lower Peak to Average Power Ratio – Flexible resource size in frequency – Contiguous resource allocation required – Some residual interference between users 61
Salah Eddine EL AYOUBI – October 2010
- TG
FFT
g*N
S/P
P/S
User 1
IFFT
Coding
DFT
. . .
User 1 Modulation
IDFT
h*0
User 2 h*N
IDFT
Multi-carrier … the frequency dimension E-UTRAN uplink sub-frame format – Same basic parameters as downlink – Contiguous resource allocation – Frequency hopping between slots Spectrum allocation (half sub-frame) and between sub-frames allowed for diversity. 1.25 - 20 MHz – Control only channels are Modulated allocated on both sides part of band of the band. ~ 60% – If data allocation exists control is multiplexed with data in the same resource.
Frequency
L1/L2 Control
1ms Normal Sub-frame
62
Salah Eddine EL AYOUBI – October 2010
User A
User B
Multi-carrier … the frequency dimension
frequency adaptive scheduling – Choose best time frequency resources based on channel quality feedback – Additional scheduling dimension compared to HSDPA (time only) – Reliable feedback can only be obtained for low speed users
interference coordination – Power restrictions allow for soft/adaptive frequency re-use
P(f)
2
Cell 1 f
– Gains seen in particular for varying load distributions
7
3
P(f)
Cells 2, 4, 6
1
f
6
4
P(f)
Cells 3, 5, 7 5 f
63
Salah Eddine EL AYOUBI – October 2010
Multi-antenna … the spatial dimension
MIMO increases spectrum efficiency
NTX
NRX
– Theoretical Maximum: Spectrum Eff. = min(NTX, NRX) x Single antenna Eff.
Yes but… – Additional antenna branches are costly especially on the terminal side – Achievable rates highly depend on propagation conditions – Mobile feedback required for high rates -> limitation of supported speeds
64
Different and adaptive solutions required depending on the deployment scenario (coverage vs. rate trade-off). Salah Eddine EL AYOUBI – October 2010
Multi-antenna … the spatial dimension
multi-antenna mechanisms in E-UTRAN downlink – Space diversity for improved robustness of common control channels and for users with high speed and/or low rate – Beamforming for coverage limited deployments
A) Transmit diversity -> Increased robustness
B) Beamforming -> Increased coverage
C) Spatial multiplexing -> Increased throughput
D) Multi-user beamforming (SDMA) -> Increased capacity
– Spatial multiplexing for high rates near the base station Adaptive selection of number of layers. – Spatial multiplexing of users in scenarios with high user density and low rate traffic
Only single antenna transmission considered in E-UTRAN uplink – Spatial multiplexing of users with multiple antennas at the base station receiver.
65
Salah Eddine EL AYOUBI – October 2010
Multi-antenna … the spatial dimension
Transmit diversity – Space diversity takes advantage of spatial A) Transmit diversity de-correlation to mitigate fast fading -> Increased robustness – Large antenna spacing or cross-polarized setups are preferred. – Receive diversity does not require a specific scheme and always gives gain, even for high fading correlation (>3dB for 2 Ant). – Transmit diversity schemes rely on redundancy transmitted from the different antennas and can work with single receive antenna. – Low correlation between antennas is essential since no power gain is achievable at the transmitter (power is distributed over antennas). – Space-Time Block Codes (or Space-Frequency Block Codes with OFDM) are low complex transmit diversity schemes.
66
Salah Eddine EL AYOUBI – October 2010
Multi-antenna … the spatial dimension
Transmit diversity in E-UTRAN – Transmit diversity can be applied to all downlink A) Transmit diversity channels in E-UTRAN (broadcast, control, data) -> Increased robustness – Basic scheme is Space Frequency Block Coding (SFBC) Orthogonal encoding avoids interference between symbols and simplifies the receiver (linear receiver is sufficient)
– Transmit diversity can be combined with multi-layer transmission using so-called cyclic delay diversity (CDD). 67
Salah Eddine EL AYOUBI – October 2010
Multi-antenna … the spatial dimension
Beamforming – Beamforming concentrates energy to increase transmission rates at cell edge. B) Beamforming -> Increased coverage – Small antenna spacing and spatially correlated fading (small angle spreads) are preferred. – Channel state information (CSI) needed at transmitter (at least Direction(s) Of Arrival, DOA) – CSI can be obtained from uplink estimations (in particular in TDD systems) or from terminal feedback (costly). – Beamformed dedicated (user specific pilots) are needed to enable channel estimation at the terminal. – Broadcast and control channels cannot be beamformed. – DL Coverage is determined by these channels – Common reference signals are needed for broadcast & control.
– Calibration of antenna arrays is a practical technical challenge. 68
Salah Eddine EL AYOUBI – October 2010
Multi-antenna … the spatial dimension
Beamforming illustrated:
Single-user approach – maximisation of the SNR. – implicit interference reduction – knowledge of user DoA
Multi-user approach – Maximisation of the SINR. – Explicit interference reduction – Knowledge of all DoAs
Antenna: ULA, M = 8 Users: 2 (Car: 1 DOA/ Phone: 2 DOAs) 69
Salah Eddine EL AYOUBI – October 2010
Multi-antenna … the spatial dimension
Beamforming in E-UTRAN – Dedicated reference signals for a B) Beamforming single stream are supported. -> Increased coverage – Terminal estimates CQI from common reference signals, BS estimates beamforming gain for link adaptation. – BF gain is approximately 10log(M) dB – Codebook based pre-coding (~fixed beams) is supported and can also be combined with multi-layer transmission. – Mobile feeds back index of preferred pre-coding vector and can obtain channel estimates from common pilots multiplied by known pre-coding vector.
70
Salah Eddine EL AYOUBI – October 2010
Multi-antenna … the spatial dimension
Spatial multiplexing C) Spatial multiplexing
– Exploits good channel conditions to -> Increased throughput transmit via parallel layers. – Prefers rich scattering and un-correlated fading (large antenna spacing's or cross-polarized setups) – Transmitter scheme: FEC
N spatial layers
FEC
Mod. Mod.
CQI feedback for link adaptation
Precoding wN
M Txantennas
Precoding w1
Feedback of pre-coding vector index
– Receiver needs as many antennas as layers to be received. 71
Salah Eddine EL AYOUBI – October 2010
Multi-antenna … the spatial dimension
Spatial multiplexing receiver
C) Spatial multiplexing -> Increased throughput
– Serial Interference Cancellation (SIC) receiver: Detect first codeword, if CRC correct re-generate interference contribution and subtract before decoding second codeword, …
Space Time LMMSE
Symbol detection
Source: A. Saadani
Serial Interference Cancellation 72
Salah Eddine EL AYOUBI – October 2010
Multi-antenna … the spatial dimension
Spatial mutliplexing in E-UTRAN – Up to 2 codewords per user. – Coverage vs. Rate trade-off:
C) Spatial multiplexing -> Increased throughput
Source: Ericsson 73
Salah Eddine EL AYOUBI – October 2010
Multi-antenna … the spatial dimension
Multi-user MIMO – Different layers can be transmitted D) Multi-user beamforming (SDMA) to different users in downlink. -> Increased capacity – E-UTRAN uses same codebook as for single user multiplexing. – Challenge to estimate CQI at terminal, since potential interference of other users is not known in advance. – Multi-user MIMO can enhance capacity in the uplink. – Transparent to the UE, only separable reference signals need to be used. – Multi-user MIMO is only useful for medium/low rate services with very high user densities. – Control signaling will become the limiting factor for user capacity. 74
Salah Eddine EL AYOUBI – October 2010
Multi-layer – packet oriented radio
Fast packet scheduling in E-UTRAN – Reduced transmission interval of 1ms – Fast packet scheduling – Fast link adaptation and cross-layer design
Benefits – Reduced latency – Performance gains from adaptive configuration and multi-user diversity
Yes but… – Amount of signaling is increased -> higher overheads – Robustness to feedback errors and high velocities
75
Salah Eddine EL AYOUBI – October 2010
Multi-layer – packet oriented radio
Cross-layer design (Layer 1 – Layer 2) Fast fading
Transmission time
user throughput
~ Fixed ressource allocation
Achievable Throughput
global throughput
~ Achievable Throughput
Time
User 1
good
Multi-user diversity gain bad
76
Circuit oriented and layered design
User 2
Time
Fast fading
User 1
Intelligent scheduling with feedback
Packet oriented and cross layer design
User 2
Usage of terminal feedback for resource allocation and phy-layer configuration
Cross-layer mechanisms already implemented in HSDPA.
Extension to frequency adaptive scheduling and adaptive MIMO transmission Salah Eddine EL AYOUBI – October 2010
Uplink power control in E-UTRAN Data
Interference coordination interference
Intra-cell power control To control Received Data Quality
77
Inter-cell power control To control Received Interference
Combination of open loop power control with closed loop adjustments
Closed loop updates are send les frequently than for UMTS (<200 Hz) No intra-cell interference between users
Regular updates are send for power control of control channels
A periodic updates are send together with the UL scheduling grant
Inter-cell interference coordination is achieved via NodeB-NodeB communication. Salah Eddine EL AYOUBI – October 2010
Uplink power control in E-UTRAN
Basic formula implemented in the terminal: P = min ( Pmax , 10 log M + Po + α x PL + delta_mcs + f(delta_i))
78
Po : UE specific offset
α : Fractional Path-Loss compensation (cell specific)
M : the number of assigned RBs in the uplink grand (only for data)
delta_mcs : MCS specific correction
delta_i : cumulative or absolute correction value per UE signalled in the UL grant (data channels) or periodically (control channels)
Discussion still ongoing in particular on the interactions with interecell coordination
Salah Eddine EL AYOUBI – October 2010
outline: LTE
79
physical layer
throughput calculations
capacity calculations
use case: mobile TV
Salah Eddine EL AYOUBI – October 2010
what is interference in OFDMA?
80
no intra-cell interference
inter-cell interference is due to collisions between chunks
Salah Eddine EL AYOUBI – October 2010
interference calculations
Exercise
81
Salah Eddine EL AYOUBI – October 2010
link budget for throughput calculations
link budget objective maximum distance between a user and its serving base station while guaranteeing a given quality of service
equipment parameters
82
propagation model
Salah Eddine EL AYOUBI – October 2010
received signals
SINR
throughput
link level curves
provide throughput vs SNR curves according to: – multiple antenna use (SISO, MIMO) – channel model (AWGN, Vehicular A, ..) – speed
83
Salah Eddine EL AYOUBI – October 2010
main output
stand-alone user throughput as a function of the distance to the base station
Max throughput DL Cell Throughput versus Distance 18000 DL Cell Throughput (Kbps)
16000 14000 12000 10000 8000 6000 4000 2000 0 0,000
84
Salah Eddine EL AYOUBI – October 2010
0,050
0,100 0,150 Distance (Km)
0,200
0,250
Throughput @ cell edge
application: impact of some design parameters inter-site distance impact on DL average cell throughput: – when the cell is larger, a larger proportion of users is at cell edge
neighboring cell load impact on DL average cell throughput:
DL average cell throughput vs ISD 6.0 DL average cell throughput (Mbps)
5.5 5.0 4.5 4.0 3.5 0
2
3 4 Inter-site distance (km)
5
6
DL average cell throughput vs DL load 16.0 DL average cell throughput (Mbps)
– when the load of neighboring cells increases, inter-cell interference increases
1
14.0 12.0 10.0 8.0 6.0 4.0 0%
85
Salah Eddine EL AYOUBI – October 2010
20%
40% DL load (%)
60%
80%
outline: LTE
86
physical layer
throughput calculations
capacity calculations
use case: mobile TV
Salah Eddine EL AYOUBI – October 2010
how can link budget help capacity analysis?
link budget gives the throughput vs distance: – throughput depends on position
cell can be decomposed into rings: – To simplify analysis – Homogeneous throughput in each ring DL Cell Throughput versus Distance 18000 DL Cell Throughput (Kbps)
16000 14000 12000 10000 8000 6000 4000 2000 0 0,000
87
0,050
0,100 0,150 Distance (Km)
Salah Eddine EL AYOUBI – October 2010
0,200
0,250
voice traffic: multi-Erlang analysis
Consider voice traffic – Calls arrive with Poisson rate λ – Stay in communication for an average time T=3min – Require each 20 Kbps, or are blocked otherwise.
Example: 2 rings – 1 Mbps for cell center, 500 Kbps for cell edge – One cell center (edge) user occupies 2% (4%) of the resources – Admission control constraint: 2*Kcenter+4*Kedge<100
Multi-Erlang analysis can be used to assess capacity: – Several classes corresponding to the number of rings – Gives blocking rates
88
Salah Eddine EL AYOUBI – October 2010
Best effort traffic: average cell throughput
Consider best effort traffic – Calls arrive with Poisson rate λ – Stay connected until transmitting a file of average size 1 Mbits
Example: 2 rings – 1 Mbps for cell center, 500 Kbps for cell edge – One cell center (edge) user stays in average 1 second (2 seconds) in the cell until transmitting its file – the time necessary for the two users to transmit their files is 1+2=3 seconds – Within these three seconds, the volume of data transferred is equal to 2 files= 2 Mbit. – The average throughput of the cell is then: T=2 Mbit/3 second=667 Kbps
89
Salah Eddine EL AYOUBI – October 2010
Best effort traffic: Arithmetic versus harmonic mean
The arithmetic mean of the throughput is: Tarith=(1 Mbps+0.5 Mbps)/2=750 Kbps
This is different from the average throughput calculated previously.
However, this corresponds to the harmonic mean: Tharm={[(1 Mbps)-1+(0.5 Mbps)-1]/2}-1=667 Kbps
90
This harmonic mean gives larger weights for cell edge users as they stay longer in the cell
The harmonic mean is convenient to measure the cell throughput
Salah Eddine EL AYOUBI – October 2010
Best effort traffic: Harmonic mean calculations DL Cell Throughput versus Distance 18000 DL Cell Throughput (Kbps)
16000 14000 12000 10000 8000 6000 4000 2000 0 0,000
91
0,050
0,100 0,150 Distance (Km)
0,200
0,250
Represents the maximal traffic that can be carried by the cell.
Used since the paper of Bonald el al., 2003.
Salah Eddine EL AYOUBI – October 2010
Best effort traffic: Processor sharing
Objective: – Estimate QoS for a given traffic
Data users share the remaing resources – not used by streaming and voice ones (priority to streaming/voice) – Fair in time, but not fair in throughput
Processor sharing analysis can be used to assess capacity: – Several classes corresponding to the number of rings – Gives average individual throughput at each position of the cell.
92
Salah Eddine EL AYOUBI – October 2010
general model
predict the QoS based on marketing traffic forecasts.
determine the number of resources needed to ensure a target QoS.
streaming traffic
streaming QoS multi-Erlang
data traffic category distribution
PS data QoS
throughput pdf (link budget) 93
Salah Eddine EL AYOUBI – October 2010
outline: LTE
94
physical layer
throughput calculations
capacity calculations
use case: mobile TV
Salah Eddine EL AYOUBI – October 2010
Use case: TV traffic
mobile TV traffic expected to explode
unicast too greedy in resources: spectrum resources
TV traffic evolution 6
8 7
Erlang
6 5
×15
4 3 2 1 0 2009
95
carriers of 5 MHz
5
4
3
2
1
2010
2011
2012
Salah Eddine EL AYOUBI – October 2010
2013
0 2009
2010
2011
2012
2013
broadcast solution
Point to Multipoint is the solution
adapt to radio conditions QPSK 1/2 16QAM 1/2 16QAM 3/4 64QAM 3/4
96
transmit with QPSK ½
transmit with 16QAM ½
advantage: simple
advantage: optimal
drawback: suboptimal
drawback: needs feedback
Salah Eddine EL AYOUBI – October 2010
total broadcast: Single Frequency Network
if every body is watching TV – why not cooperating all base stations?
Interference is seen as a multipath propagation
drawback: tight synchronization between cells is needed
97
Salah Eddine EL AYOUBI – October 2010
Delay and multipath impact
Weight function for the constructive portion of a received SFN signal:
w(delay ) = 1 if 0 ≤ delay ≤ TCP w(delay ) =
TCP + Tu − delay Tu
if TCP < delay < TCP + Tu
w(delay ) = 0 if TCP + Tu < delay w(delay) 1
Tcp
98
Tu+Tcp
delay
Take into consideration of multipath propagation in the calculation
Salah Eddine EL AYOUBI – October 2010
SINR calculation
Based on the weight function w() and the multipath table, the following equation for calculating average SINR per subcarrier for a MBSFN user M(r, θ) in cell 0 is derived: DL N 6
∑∑
SINRsubc ( M ) = where:
99
w( d ( M , j ) + dm ( p )) × rm ( p) × Psubc
j = 0 p =1
N
6
∑∑
DL (1 − w( d ( M , j ) + dm ( p )) ) × rm ( p ) × Psubc
j = 0 p =1
Shadowing is also to be considered. Salah Eddine EL AYOUBI – October 2010
PLDL M,j PLDL M,j
+ N th
SFN parameters LTE MBSFN FDD inputs
FDD Case 1
FDD Case 2
FDD Case 3
Carrier spacing (∆f)
15 kHz
15 kHz
7.5 kHz
Symbols number per subframe
14
12
6
Subcarriers number per RB (Resource Block)
12
12
24
MBSFN reference symbols number per RB
18
18
18
Cyclic prefix duration (TCP)
4.69 μs TCP = 144×Ts (for OFDM symbol #1 to #6) Ts = 1/ (2048 × ∆f)
16.67 μs TCP-e = 512×Ts (for OFDM symbol #0 to #5) Ts = 1/ (2048 × ∆f)
66.67 μs TCP-low = 1024×Ts (for OFDM symbol #0 to #2) Ts = 1/ (2048 × ∆f)
Extended Vehicular A (EVA)
100
relative delay (ns)
Relative Mean Power[dB]
0
0
1
30
-1.5
0.71
150
-1.4
0.72
310
-3.6
0.44
370
-0.6
0.87
710
-9.1
0.12
1090
-7.0
0.20
1730
-12.0
0.06
2510
-16.9
0.02
Salah Eddine EL AYOUBI – October 2010
Power ratio [lin]
comparing TV deployment strategies
when TV traffic increases – unicasting it will be disastrous on other services… – single-cell broadcast (PtM) with modulation adaptation is a good intermediate solution – total broadcast (MBSFN) is the best solution…
101
Salah Eddine EL AYOUBI – October 2010
outline
objective: ensuring QoS in mobile networks
dimensioning for ensuring coverage
dimensioning for ensuring capacity – GSM – UMTS – LTE – just before LTE: HSDPA – after LTE: LTE-A
102
Salah Eddine EL AYOUBI – October 2010
hybrid CDMA/TDMA access CDMA radio
reduced inter-cell interference residual intra-cell interference 103
Salah Eddine EL AYOUBI – October 2010
no CDMA in the same cell – TDMA
how it works
104
scrambling code is used to reduce inter-cell interference
within the same cell, channelization code is used to separate signalling, UMTS and HSDPA signals.
HSDPA users share the same code in time
when there is at least one HSDPA user in the cell, the base station emits at maximal power
Salah Eddine EL AYOUBI – October 2010
Radio model 1.
The SINR for a mobile depends on the distance r0 from the BS, as intercell interference increases at cell edge.
2.
To simplify the problem, divide the cell into concentric rings: A mobile is thus charcterized by its service and its position in the cell.
3.
Calculate R99 power and deduce the power remaining for HSDPA, as shown below. Pmax
HSDPA
Variable rate
data R99
Constant rate
voice R99 Common channels 105
Salah Eddine EL AYOUBI – October 2010
admission control for R99
power of base station limited by Pmax
wa can show that admission control constraint remains the same as for pure R99 systems: n
∑ (αP
max
+ χ Pmax Fi + N 0 qi )(
i =1
C
∑β M c
c =1
intra-cell interference noise intra-cell interference
106
Salah Eddine EL AYOUBI – October 2010
i ,c )
≤ Pmax − PCom
HSDPA throughput
4000
throughput (Kbps)
3500 3000 DCH=0 2500
DCH=20%
2000
DCH=40% DCH=60%
1500
DCH=65% 1000 500 0 0,040
0,060
0,080
0,100
0,120
0,140
distance to base station (Km)
107
Salah Eddine EL AYOUBI – October 2010
0,160
0,180
UMTS/HSDPA model
UMTS traffic
multi-Erlang
UMTS QoS
PDCH
108
streaming traffic
multi-Erlang
streaming QoS
data traffic
PS
data QoS
Salah Eddine EL AYOUBI – October 2010
outline
objective: ensuring QoS in mobile networks
dimensioning for ensuring coverage
dimensioning for ensuring capacity – GSM – UMTS – LTE – just before LTE: HSDPA – after LTE: LTE-A
109
Salah Eddine EL AYOUBI – October 2010
fast…
relays – increases useful signal
open issues: – choosing the relay type (AF, DF) – dimensioning of links
110
Salah Eddine EL AYOUBI – October 2010
fast…
relays
coordinated multipoint (CoMP)
– increases useful signal
open issues: – choosing the relay type (AF, DF) – dimensioning of links
111
Salah Eddine EL AYOUBI – October 2010
– increases useful signal – decreases interference
open issues: – choosing the CoMP type – dimensioning of backhaul