Recent Results in Wireless Systems: From Smart Roaming to the Use of - - PowerPoint PPT Presentation

Recent Results in Wireless Systems: From Smart Roaming to the Use of Wide Channels in 802.11ac

Catherine Rosenberg

Canada Research Chair in the Future Internet CISCO Research Chair in 5G Systems

This work was done in collaboration with B. Venkitesh, S. Malekmohammadi, R. Stanica

1

Outline

• WiFi Networks: Channel Allocation in WiFi.ac
• Cellular Networks: Smart Roaming

2

IEEE 802.11 ac (WiFi 5) Features and Parametrization

• Features:
• Channel bonding (20/40/80/160 MHz)
• Using more MIMO spatial streams (8)
• Using denser modulation schemes (256 QAM)
• The allocation of the resource and the parametrization are critical, especially in dense

environments, to provide good performance:

• Channel (size and id)
• Transmission power of each AP
• Carrier sensing threshold (CST) used by each AP
• Research questions: When is it better to use wider channels? When is it better to use

lower power?

3

IEEE 802.11 ac channel allocation in the 5 GHz band

WiFi ac

Current Practice in Industry

Power: 2 main trends

• Fixing the transmission power for each AP to a high fixed value (usually the maximum power

defined in the standard)

• Assigning dynamic power values to APs obtained from a power management algorithm. These

values are always greater than a lower bound PTPC that we compute. CST:

• Usually, a default value is used (e.g. -82 dBm).

Channel and Bandwidth:

• Having fixed the other two parameters, there are different ways to allocate channels, the general

trend being to allocate narrower channels as the network density increases. Most of the works in the literature study the effect of only one of the channel bandwidth, transmission power and CST parameters on the networks performance, while the other two are fixed. There are also some proposed algorithms for power management and channel allocation, but the two are performed independently from each other by a central controller.

4

WiFi ac

Our Claim

There is need to study the impact of the three parameters together. Much gain can be obtained by using wide channels with low power (instead of using narrow channels with high power)

5

WiFi ac

Benchmarks

• Benchmark 1: use maximum power, default CST and select smaller size channels

when the network is dense (By the way when is a network deemed dense is not typically defined).

• Benchmark 2: a solution proposed by CISCO that allocates power and channel

dynamically but with a power that is always greater than a value that we computed.

6

WiFi ac

Framework

7

• Topology: we consider an enterprise WiFi network with 64 identical APs located on four

floors of a building. In each room, there is one AP associated with four users.

WiFi ac

• We use NS3 to simulate realistic WiFi networks of different densities.

We only consider downlink transmissions, as this is the main use case in WiFi networks.

• To simplify the search space and to have only one 160 MHz channel, we
• nly consider one band of 160 MHz. We also use the same channel

width, power and CST for all APs (except when otherwise noted).

L L

Framework

8

Channel Propagation and Power:

• We use the NS3 Log Distance Propagation Loss Model.
• There is a maximum power of 23 dBm (200 mW). Consequently, when using a channel width of 20, 40,

80 and 160 MHz, the maximum power that can be used over each 20 MHz sub-channel (P20

t) is equal to

23, 20, 17 and 14 dBm, respectively. Coverage and Transmission Power:

• We define the coverage range of an AP as the closest distance at which a receiver is not able to decode

the transmitted signal. We can compute it easily given P20

t or alternatively we can find the minimum

power P20

t,min for a given range R. For our topology: P20 t,min = 30 Log10 (((L+5)/√2) +5) - 47.33

• The lower L, the higher the deployment density and the lower the corresponding P20

t,min

Traffic and Overall Performance Metric:

• We assume a full buffer UDP traffic with a uniform arrival rate of 230 Mbps.
• The overall performance metric that we use is the geometric mean (GM) of the throughput of all the

users: GMtot = N√ (∏uλu)

WiFi ac

Results

9

• Same channel width,

power and CST for all APs.

• L= 40m (low density).
• The best performance is

≈ 84 Mbps, obtained for W = 160 MHz and P20

t ≈ 6

dBm.

WiFi ac

Benchmark 1 Benchmark 2 Proposed Method P20

t,min

L= 40m, 35 Mbps (W = 160 MHz) (P20

t = 14 dBm)

53 Mbps (W = 160 MHz) (P20

t = 11 dBm)

84 Mbps (W = 160 MHz) (P20

t = 6 dBm)

67.6 Mbps (W=160 MHz) (P20

t = -0.347 dBm)

Results

10

Benchmark 1 Benchmark 2 Proposed Method P20

t,min

L= 25m, 11.87 Mbps (W = 160 MHz) (P20

t = 14 dBm)

29.49 Mbps (W = 160 MHz) (P20

t = 8 dBm)

53.52 Mbps (W = 160 MHz) P20

t = 0 dBm

44.7 Mbps (W = 160 MHz) P20

t = -4.77 dBm

WiFi ac

v v

• Same channel width,

power and CST for all APs.

• L= 25m (medium

density):

Results

11

• Same channel width,

power and CST for all APs.

• L= 15m (high density)

WiFi ac

v v v Benchmark 1 Benchmark 2 Proposed Method P20

t,min

L= 15m, 5 Mbps (W = 20 MHz) (P20

t = 23 dBm)

8.72 Mbps (W = 20 MHz) (P20

t = 10 dBm)

25.98 Mbps (W = 160 MHz) P20

t = -5 dBm

22.2 Mbps (W = 160 MHz) P20

t = -8.87 dBm

v

• We have used so far the same channel width, power and CST for all Aps.
• In that case, the best power value depends among other things on the density of the
• network. The denser the network, the lower the best power.
• Using the largest channel width with lower power (even with P20

t,min) is significantly

more effective than using a narrow channel width with any power.

• However, in high density deployments, there is an unfairness among APs in the

network, even when using the best power value!

12

Results

L=25m L=15m

WiFi ac

To alleviate this unfairness problem, we have tried to adjust the power and CST differently for poor APs:

• Increasing their CST so that they can get access to the shared channel more easily
• Increasing their transmission power values slightly (the other APs keep the power value

determined earlier; i.e., for the case with equal parameters) so that they can combat the higher interference that they experience  It is better to increase the CST of poor APs instead of increasing their power values

13

Results

Cpoor + Ppoor+ GMtot AMtot GMmin none none 22.8 29.51 3.92 3 6 none none 23.18 24.24 29.72 30.05 9.23 9.47 none 3 22.27 29.14 7.21 3 3 3 6 21.04 21.14 28.58 28.56 7.49 7.52

WiFi ac

Results (GM in Mb/s)

Performance (GMtot) of the proposed and benchmark methods for random realizations of users and APs 14

Benchmark 1 Benchmark 2 Proposed Method L= 15m, realization 0 L= 25m, realization 0 L= 40m, realization 0 5 11.87 35 8.72 29.49 53 25.98 53.52 84 L= 15m, realization 1 L= 25m, realization 1 4.65 12.96 9.66 26.47 27.91 52.52 L= 15m, realization 2 L= 25m, realization 2 4.85 13.67 8.97 25.03 27.85 52.71

WiFi ac

Conclusion

Densely deployed networks are characterized by many APs serving small coverage areas. So it is not necessary for the APs to use high values of power. Instead, allocate low values of power to them and use wide channels.

15

WiFi ac

Outline

• WiFi Networks: Channel Allocation in WiFi.ac
• Cellular Networks: Smart Roaming

16

Introduction

• 5G is emerging to meet the ever increasing demand for mobile applications:
• 4G on steroids +
• Critical applications +
• IoT

1G

2G

3G

4G

1000x Mobile Data Volumes 10x-100x Connected Devices 5x Lower Latency 10x-100x End User Data Rates 10x Battery Life for Lower Power Devices

5G

[1] Qualcomm Technologies, Inc. , Leading the world to 5G, February 2016

Smart Roaming

17

5G in a nutshell

• 5G wireless networks promise dramatic improvements in terms of data rate,

spectral efficiency, user experience, mobility, latency, and connectivity over current state of the art 4G networks.

• Achieving the promise of 5G will require a dramatic rethink of wireless

networks.

• Everyone is talking about the PHY technologies that will enable 5G, i.e.,

massive MIMO, mmwave, full duplex, etc.

• However, 5G will also need advances at the higher layers, e.g., at the MAC

and transport layers and old concepts need to be revisited: e.g., spectrum sharing and new concepts need to be adapted to wireless, e.g., slicing.

18

Smart Roaming

Resource Management: The Focus of My Work

• We consider a multi-cell OFDMA system comprising several base-stations

(BTS).

• The resources are: channels, power (on the downlink), time.
• The resource management processes are:
• Channel allocation to the BTSs;
• User association;
• User scheduling to allocate locally time, channel and power on a fast time scale.

19

Uplink Downlink

Smart Roaming

Our focus: coordination and cooperation (1)

• Cellular networks consist of multiple base-stations, possibly heterogeneous.
• The online operation of a BTS is affected by its neighboring BTSs because

mainly of inter-cell interference (ICI).

• Coordination between BTSs can provide significantly improved performance.

This can be achieved through Cloud-RANs.

• For example, user scheduling, resource allocation, beamforming, or mobility

management (e.g., BTS handover) could benefit from BTS coordination at the cost of information exchange and possibly relocating some of these processes in the network.

20

Smart Roaming

Our focus: coordination and cooperation (2)

• Operators can cooperate to take better advantage of their resources.
• They could put in common some of their resources (some already share

towers), they could share license band (a very interesting topic), virtualized base stations or exchange users.

• We will focus on a cooperation based on what we call smart roaming where

users can change operators for other reasons than coverage, i.e., for performance reasons.

21

Smart Roaming

Smart Roaming

• Smart Roaming (SR) is about allowing users to be shared between operators covering the

same region for efficiency reasons.

• SR is an inter-operator cooperation method that
• Is cost effective,
• Simple, scalable and
• Uses existing roaming mechanisms
• Motivation
• Base stations of different operators are seldom co-located, even if they are, sectors are

rarely aligned.

• SR leverages spatial diversity, i.e., users that are at the edge of one operator might be

better off with another one.

22

Base station locations from different operators in Toronto Smart Roaming

Research Questions

• How much gain can be achieved by sharing users between operators?
• What are the factors that affect the gain?
• How to deal with operator heterogeneity to avoid that a large operator cross-

subsidize a smaller one?

• Operators with same/different number of users.
• Operators with same/different bandwidth.
• Operators with same/different number of Base stations in an area.
• How to implement such schemes in an online fashion ? In a way we can
• Control the user sharing
• Avoid cross-subsidising
• Limit the sharing of information between operators

23

Smart Roaming

Methodology

• Smart Roaming is essentially a User Association (UA) problem but with a twist

(equal gain).

• We start with a static case, where we propose a model to evaluate the performance
• f SR on the downlink that enforces equal gain to avoid cross subsidies.
• We compare its performance with a benchmark, where each operator works

independently and show, through extensive computations, that the gains can be significant.

• We then move to a dynamic setting and propose a solution based on user

scheduling even if we originally thought that the natural solution would be UA- based!

24

Smart Roaming

System Model

Consider a region where several operators have deployed their sectored BTSs.

• Q - set of operators
• Zq - set of sectors of operator q
• We focus on the downlink case
• Operator q has been licensed Mq sub-channels, each of bandwidth b.
• Each user equipment (UE) can associate with only one sector and all the users are greedy,

i.e., they want the best possible throughput.

• We select proportional fairness as our objective function in all the optimization

frameworks.

25

Smart Roaming

Snapshot/Static Model

• For a given system of operators Q, set of sectors Z and Mq’s, we begin by

considering a snapshot model.

• A snapshot or a realization ω is characterized by set of users and their channel

gains to all the sectors.

• We generate a set Ω of realizations and study the performance of the different

scenarios averaged over the entire set.

• We first introduce the benchmark, called “No Roaming” (NR) and then focus on

the smart roaming case, called “Smart Roaming/ Equal Gain” (SR/EG).

26

Smart Roaming

Benchmark in Static Case

• We call our benchmark “No Roaming”.
• For operator q and a given realization, we define the optimization model as follows (joint

centralized UA and scheduling problem, one problem per operator),

27

Maximizing GM of Throughput Scheduling variables

Smart Roaming

UA variables

Smart Roaming/Equal Gain (SR/EG)

• For a given realization and per operator benchmark, SR/EG can be written as follows (joint

centralized UA and scheduling problem, one single problem involving all operators)

28

Maximizing GM of Operator 1 Equal Gain Constraint

Smart Roaming

This is a large problem that requires the computing of the individual operator performance Γq

(0)

Numerical Results- 2 Operators

• We consider a wrap around model of deployment to avoid corner

cases.

• For two operator system, a network topology is characterized by two

parameters (Ѳ, d)

• 7 BTSs for each operator
• Each sector of second operator is rotated by Ѳ
• BTSs of operator 2 is shifted by d meters
• If d=0, the BTSs are co-located.

29

Smart Roaming

Numerical Results- 2 Operators

• We consider two cases

1. Case 1- Same number of sub channels (M1 = M2 = 99) 2. Case 2- Different number of sub channels (M1 = 99, M2 = 66)

• We evaluate SR/EG wrt NR for numerous topologies via the per operator gain,

30

Benchmark Performance SR/EG Performance

Smart Roaming

Numerical Results- 2 Operators

31

Case 1: N1=140, N2=105 Case 2: N1=140, N2=140

The maximum gain in GM throughput is 35% and in sum throughput is 29%

Smart Roaming

Numerical Results- 3 Operators

• We consider
• 7 BTSs for each operator.
• Each sector of operator 2 is rotated by Ѱ wrt to operator 1.
• Each sector of operator 3 is rotated by Ѱ wrt to operator 2.
• We consider only co-located BTSs
• We take:
• M1 = 99, M2 = M3 = 66
• N1=N2=N3=N
• Ψ ϵ {0,30,35,40,60}

32

Smart Roaming

Numerical Results- 3 Operators

33

Gains for SR/EG in three operator case

• Max GM gain of 53%
• Corresponding gain in

sum throughput is 43%

Smart Roaming

Static results show that SR has potential !

Heuristic Scheduler

• Our extensive computations have shown that the optimal scheduler is not round robin (RR), but

provides,

• Quasi equal time to all users of same operator.
• Different time to users of other operator.
• Based on this, we design a heuristic scheduler, “weighted round robin” (WRR) for a two
• perator system

𝑜1𝑘𝛽1𝑘 + 𝑜2𝑘𝛽2𝑘 = 1

𝛽1𝑘 𝛽2𝑘 = 𝑑 = 𝑎1 𝑁1𝑂2 𝑂1 𝑎2 𝑁2

where 𝑜𝑟𝑘: Number of users of operator q associated with sector j. 𝛽𝑟𝑘: Fraction of time given to user of operator q in sector j.

• Note that the ratio c is the same for each BTS of an operator.
• This can be generalized to any number of operators.
• We compare the weights of WRR to the weights of optimal scheduler over many realizations

(the UA being the optimal one) and find excellent match.

34

Smart Roaming

Summary - Static Case

• We have proposed a snapshot model for evaluating the performance of SR in the
• downlink. This model jointly optimizes UA and US
• Through extensive simulations, we show that there is a significant gain in

performance by using SR

• The gains in sum throughput can be as large as 29% for the case with 2 operators

and 43% for the case with 3 operators.

• The factors that affect gains are
• Relative distances between BTSs
• Relative angles between the antennas
• We show that a heuristic weighted round robin (WRR) scheduler is quasi optimal for

2 and 3 operator systems.

35

Smart Roaming

Dynamic Case

• We assume a Poisson arrival process for users and we consider two scenarios:
• Scenario 1: The users leave the system after staying for a random exponential time: the

metric is the rate

• Scenario 2: The users leave the system after downloading a fixed file size (10 MB): the

metric is the mean delay

• We perform extensive simulations for two operator case on co-located topology

with θ = 60 deg.

• The benchmark, called No Roaming (NR) consists of:
• Each network is operated independently with RR scheduler.
• UA is best rate
• We call Full Roaming (FR) the case where a user can select freely its BTS,

irrespective of the operator and each BTS uses a simple RR scheduler. The UA is again best rate.

36

Smart Roaming

Numerical Results- NR vs FR

37

Smart Roaming

Scenario 1: Average Download rate as a function of λ2, λ1=1.25 Scenario 2: Average Download rate as a function of λ2, λ1=1.25

Full Roaming creates cross-subsidies!

We are aiming at no cross-subsidies

38

Scenario 1 Scenario 2

Smart Roaming

New Schemes

• We have proposed two online schemes to implement SR online
• 1. Scheme 1: Any arriving user is free to choose the best BTS from either operator and each

BTS in the region uses a WRR scheduler.

• 2. Scheme 2: UA is controlled to provide load sharing via a distributed algorithm that we

design and each BTS uses its legacy scheduler. This was not a success so I will skip it. We found that UA does not have the right granularity to control the system the way we want.

39

Smart Roaming

The C-RANs coordinate by sharing information periodically

• Unlike NR and FR, the proposed schemes require
• perator coordination

Scheme 1

• The principle is simple:
• Each sector uses a WRR scheduler to discriminate between users of different operators
• UA is free for all, i.e., each user can select the best sector
• To compute the weights for WRR scheduler, we use the equations from static case.
• The ratio c is a function of quantities that are constant (number of BTS, sub

channels) and that vary with time (number of users).

• The C-RANs will periodically exchange an estimate of their number of users, i.e.,

the number of users averaged over the last one minute.

• Then, each C-RAN will compute the ratio and send the latest value to its BTSs

every minute.

• UA: A user will select the best sector irrespective of the operator that offers the

best rate. To compute an estimate of the rate, it needs: c, 𝑜1𝑘,𝑜2𝑘. Each BTS sends this info in its beacon.

40

Smart Roaming

Numerical Results

41

Scenario 1: Average Download rate as a function of λ2, λ1=1.25 Scenario 2: Average Download rate as a function of λ2, λ1=1.25 Near Constant

Smart Roaming

Numerical Results- 3 Operator case

42

Scenario 1 Scenario 2 Near Constant

Smart Roaming

Conclusions

• SR can bring significant gains with little cost as long as BTS sectors are not aligned,

even when they are co-located.

• The gains in sum throughput can be as large as 29% for the case with 2 operators and

43% for the case with 3 operators.

• A simple scheme of free UA with RR scheduler leads to cross subsidies.
• The most efficient way to implement SR is by using SR1 where each BTS changes its

scheduler from RR to WRR.

• Operators may be reluctant to change their scheduler, but governing bodies like FCC

can mandate it in certain bands

• Need to investigate SR in the uplink and in tougher scenarios where operators have

different hotspots.

43

Smart Roaming