Scheduling and capacity estimation in LTE

Olav Østerbø, Telenor CD (Corporate Development)

ITC-23, September 6-8, 2011, San Francisco
Agenda
Introduction
Obtainable bitrate as function of SINR
Radio channel propagation model
Radio signal fading model
Analytical models for LTE radio network performance
Numerical examples
Conclusions

2 ITC-23, September 6-8, 2011, San Francisco

 Introduction
LTE is the new, (following up of HSPA) mobile access
technology specified by 3GPP:
• Flat all-IP architecture
• Flexible in frequency bands (700 MHz-2.6 GHz)
• Flexible carrier bandwidths (1.4, 3, 5, 10, 15 and 20
MHz)
• Increased spectrum efficiency based on OFDMA for
uplink, SC-FDMA for downlink
– Typical cell capacity (20 MHz bandwidth), 20 – 40
Mbps for downlink link and 5 – 15 Mbps for uplink
• Momentum in the industry, building on current
investments in the GSM/UMTS

3 ITC-23, September 6-8, 2011, San Francisco

Obtainable bitrate as function of SINR
Bitrate function B=B(SINR)
• Upper bound Shannon: B/f=Log2(1+SINR)
• Discrete; B/f based on CQI-table (3GPP TS 36.213) and Linear
relation between SINR[dB] and CQI-index
• Approximate/Truncated modified version of Shannon's formula:
B/f=MIN[T, C Log2(1+γSINR)]

CQI index modulation code rate x 1024 efficiency 6

0 out of range

@D
---- Shannon
1 QPSK 78 0.1523
5 ---- Modified Shannon

No rm ali sed T hro ugh pu t b it s H z

2 QPSK 120 0.2344
3 QPSK 193 0.3770 ---- LTE CQI table
4 QPSK 308 0.6016 4
5 QPSK 449 0.8770
6 QPSK 602 1.1758
7 16QAM 378 1.4766 3
8 16QAM 490 1.9141
9 16QAM 616 2.4063
2
10 64QAM 466 2.7305
11 64QAM 567 3.3223
12 64QAM 666 3.9023
1
13 64QAM 772 4.5234
14 64QAM 873 5.1152

4
15 64QAM 948

ITC-23, September 6-8, 2011, San Francisco

5.5547
0 5
@D
10
SINR dB
15 20 25
 Radio channel propagation model

PwG
Signal-to-noise ratio: SINR =
N
Pw sending power
Path loss (model): G = 10 L 10 L = C − A log10 (r ) + X t
C and A constants, Xt shadowing usually assumed to be normal
distributed with zero mean and given standard deviation
Noise N = N int + N ext sum the internal (or own-cell) noise power
and is the external (or other-cell) interference.

5 ITC-23, September 6-8, 2011, San Francisco

Radio signal fading model
SINR on the form: S t r α h (λ )
Stochastic part of SINR: S t = X ln X e
Slow fading (shadowing): Lognormal Xln
Fast fading: Rayleigh, i.e. neg. exp. distributed Xe −t −
(ln(t x )) 2
∞ ∞ 2σ 2
~ − xt 1 e
St : Suzuki distributed with CDF S su ( x) = ∫e sln (t )dt = ∫ dt
t =0 2π σ t =0
t
(ln(t x )) 2
−t − k 2σ 2
2σ 2
1 ∞ (−1) k k kσ
T
~ 1 e ln( x T )
Truncated version: S su ( x, T ) = ∫ dt = ∑ x e 2
erfc( + )
2π σ t =0
t 2 k =0 k! 2 σ 2
3 2

Lognormal 1.75 Suzuki σ = 5.0

2.5

σ = 0.2 σ = 5.0 1.5

σ = 2.0
2 HL
HL

1.25

σ = 2.0 σ =1.0
s x

s x

1.5 1

σ =1.0 0.75
σ = 0.6
1

σ = 0.6
0.5
0.5
0.25
σ = 0.2
1 2 3 4 5 1 2 3 4 5
6
x
ITC-23, September 6-8, 2011, San Francisco x
Analytical models for LTE radio network
performance
• Spectrum efficiency through the bit-rate distribution per
Recourse Block (RB) for users that are either randomly or
located at a particular distance in a cell.
• Cell throughput/capacity and fairness by taking the scheduling
into account.
– Scheduling based on metrics which depends (only) on own SINR
and distance
– Specific models for the common (basic) scheduling algorithms,
Round Robin, Proportional Fair and Max-SINR.
• Estimation of the capacity usage for GBR sources in LTE
– Non-persistent allocation, i.e. allocation every TTI to obtain GBR
rate
• Cell throughput/capacity for a mix of GBR and Non-GBR
(greedy) users

7 ITC-23, September 6-8, 2011, San Francisco

 Input parameters to numerical examples

Parameters Numerical values

Bandwidth per Resource Block 180 kHz=12x 15 kHz
Total Numbers of Resource Blocks 100 RBs
Distance-dependent path loss. (Taken from L=C +37.6log10(r),
a 3GPP document) r in kilometers and
C=28.1 dB for 2GHz

Lognormal Shadowing with standard 8 dB (in most of the cases)

deviation

Rayleigh fast fading

Noise power at the receiver -101 dBm
Total send power 46.0 dBm=(40W)
Radio signaling overhead 3/14

8 ITC-23, September 6-8, 2011, San Francisco

Mean throughput per RB as function of cell radius
0.8
Suzuki distributed fading,
0.7 2GHz frequency and σ=0dB,
@D
2dB, 5dB, 8dB, 12dB from
below.
Mea n thro ughput p er RB Mbi t s

0.6

Located at cell
0.5
edge
0.4
Random location
0.3

0.2

0.1

1 2
Distance @D
Km
3 4 5

Throughput per RB drops for large cells. Approx 0.2 Mbit/s for
2km cell and 0.05 Mbit/s at cell edge with 8dB shadowing.
9 ITC-23, September 6-8, 2011, San Francisco
 Multiuser gain as function of cell radius
80

2GHz frequency, 100 RBs,

70 fading Susuki distributed with
shadowing σ=8 dB, number of
@D

60 users =1, 2, 3, 5, 10, 25,

100 from below.
C el l ca pa ci ty M bi t s

50

40

30
-- PF
-- Max-SINR
20

-- RR
10

1 2
Distance @D
Km
3 4 5

Multiuser gain very large for Max-SINR. PF doubles cell

throughput compare to RR for cell of 2 km and 25 users.
10 ITC-23, September 6-8, 2011, San Francisco
Mean Bitrate for a user located at cell edge as
function of cell radius.
40

@D
@D
25
35

Capacityper user Mbit s

Capacityper user Mbit s

30
20

25
2-users 15
20 Max-
SINR 3-users
15 10 Max-
PF SINR
10
5
5 RR PF

16
1 2
Distance @D
Km
3 4 5

8
1 2
RR

Distance @D
Km
3 4 5
@D

@D
14 7
Capacityper user Mbit s

Capacityper user Mbit s

12 6

10 10-users
5-users
5

8 4

6 3

4 Max- 2
SINR PF Max-
2
1
RR SINR PF
1 2
Distance @D
Km
3 4 5
1 2
RR

Distance @D
Km
3 4 5

Max-SINR shows very poor cell Scheduling: RR, PF and Max-SINR

scheduling algorithm , 2GHz
edge performance frequency and 100 RBs

11 ITC-23, September 6-8, 2011, San Francisco

 Mean cell throughput for 10 users scheduled according to
PF and a GBR user
80 80

-- Non-Persistent, cell edge -- Non-Persistent, cell edge

@D
@D
70 70
-- Non-Persistent, random -- Non-Persistent, random
-- mean PF 10 users -- mean PF 10 users

Cel l c apa cit y Mbi t s

60
Cel l ca paci ty Mbi t s

60

50 50

GBR=3 Mbit/s 40
GBR=1 Mbit/s
40

30 30

20 20

10
10

0.5 1
Distance @D
Km
1.5 2 2.5
0.5 1
Distance @D
Km
1.5 2 2.5

80 80

-- Non-Persistent, cell edge -- Non-Persistent, cell edge

@D

@D
70 70
-- Non-Persistent, random -- Non-Persistent, random
-- mean PF 10 users -- mean PF 10 users
Cell capacity Mbit s

Cel l ca paci ty Mbi t s

60 60

50 50

40 40

30 30

GBR=0.3 Mbit/s GBR=0.1 Mbit/s

20 20

10 10

0.5 1
Distance @D
Km
1.5 2 2.5 0.5 1
Distance @D
Km
1.5 2 2.5

GBR rates of less than 1 Mbit/s does GBR of 3.0, 1.0, 0.3, 0.1 Mbit/s using
not reduce the overall throughput very non-persistent scheduling, for 2 GHz and
much. GBR rates larger than 1 Mbit/s 100 RB and Suzuki distributed fading
is not recommended. with std. σ=8dB.
12 ITC-23, September 6-8, 2011, San Francisco
 Conclusions

• The two most important factors for the radio performance in

LTE are fading and attenuation due to distance.
• Numerical examples for LTE downlink shows results which are
reasonable;
– In the range 25-50 Mbit/s for 1 km cell radius at 2GHz with 100
RBs .
– Multiuser gain is large for the Max-SINR algorithm but also the PF
algorithm gives relative large gain relative to plain RR.
– The Max-SINR has the weakness that it is highly unfair in its
behaviour. (Not recommended to use in real operation.)
• The usage of GBR with high rates may cause problems in LTE
due to the high demand for radio resources if users have low
SINR i.e. at cell edge.
– GBR rate limited to at most 1 Mbit/s per user?

13 ITC-23, September 6-8, 2011, San Francisco