QoS-aware Antenna Grouping and Cross-layer Scheduling for mmWave - - PowerPoint PPT Presentation

QoS-aware Antenna Grouping and Cross-layer Scheduling for mmWave Massive MU-MIMO [1]

[1]

• C. Bocanegra, S. Rodrigo, Z. Li, A. Cabellos, E. Alarcon and K. R. Chowdhury, “Qos-aware Antenna Grouping and Cross-layer Scheduling

for mmWave Massive MU-MIMO,” IEEE JSAC, Submitted Jan. 2020 (Under revision) Code.

> <

• 5. Performance evaluation

How much better off are we with our approach?

2

• 1. Introduce millimeter waves (mmWave) as a key technology

Why is mmWave interesting? How is it different from other bands

• 3. Our cross-layer approach for MU-MIMO in mmWave

Leverage upper-layer information to best deal with PHY-layer limitations

• 4. MATLAB Toolboxes that made it possible

Communications, Optimization and WLAN toolboxes

• 2. mmWave and MU-MIMO in the 5G and 802.11ad/ay standards

How are this standards going to operate in the band?

OBJECTIVES OF THE TALK

> <

3

• 1. Introduction to mmWaves – key factors
• 2. Beamforming to overcome losses in mmWaves
• 3. RF-antenna interconnections in mmWave
• 4. Beamforming in the 802.11ad standard
• 5. Our cross-layer approach – an overview
• 6. Internet traffic – generating realistic flows
• 7. HELB – heuristically enhanced LCMB beamforming – OPTIMIZATION TOOLBOX
• 8. The mmWave channel – COMMUNICATIONS TOOLBOX
• 9. Emulating the mmWave link – WLAN TOOLBOX + PHASED ARRAY TOOLBOX

10.Performance evaluation 11.MATLAB in the project 12.Conclusions

AGENDA

> <

[1] M. A. Rojavin and M. C. Ziskin, “Medical application of millimeter waves,” Q J Med 1998, pp. 57-66, vol. 91, 1998 [2] A. Cardama, Ll. Jofre, J. M. Rius, J. Romeu, S. Blanch and M. Fernando, “Antennas,” Edicions UPC, 2002 [3] Y. Niu, Y. Li, D. Jin, L. Su, A. V. Vasilakos, “A Survey of Millimeter Wave (mmWave) Communications for 5G: Opportunities and Challenges”, arXiv:1502.07228, submitted in February 2015 4

Wider bandwidth allows higher resollution for X-

• Rays. No ionizing

property [3] Neuron communication at 42, 53 and 61 GHz. Drug withdrawal, diabetes, cáncer, etc. [1] The higher radar resolution allows for tracking the steam of the missiles better [3] Atacama large mm Array (ALMA). A millimeter wave radio telescope in Chile [2]

INTRO – MMWAVE APPLICATIONS BEYOND COMMS

> <

INTRO – WHY MMWAVE?

5

Cellular (5G) Unlicensed (802.11ad/ay)

Advantage of mmWave

• Wider bandwidth.
• Low interference, unexploited band.

Bad

• Channel losses due to a higher carrier

frequency, and molecular absorption (60GHz).

Challenges:

• Limited coverage: ~200 meters
• Beamforming is required to deal with losses.
• Higher bandwidth, requires higher sampling

rate.

• Higher sampling rate requires higher power.

Underusage of the wireless spectrum, confined in the sub- 6GHz band Proposed

> <

INTRO – BEAMFORMING AS A KEY ENABLER

6

Analog beamforming Digital beamforming Hybrid beamforming

PRO: Low implementation complexity and costs. CONS: The resolution of the phase shifters is limited. Thus, inter-user interference is higher. PRO: Offers the optimum

• beamforming. Can create large

number of beams. CONS: Requires a large number of RF-chains, high power consumption. PRO: Reduced power consumption given with less RF-chains. RISK: Its performance depends upon the sub-array connectivity.

> <

7

RF CHAIN TO ANTENA – FLEXIBILITY ON PHY

Fully connected Phase and switch Fully connected Phase shifters Switching network Partially connected Phase shifters Flexible Partially connected

A mmWave arrays is characterized by the phase shifters and/or switches, offering different connectivity levels. The sub-array architecture selection depends upon power consumption, price, and area budget. For example, phase shifters allow for a more flexible design than switches, but their power consumption is higher.

[1] Shahar Stein Ioushua, Yonina C. Eldar, “Hybrid Analog-Digital Beamforming for Massive MIMO Systems,” arXiv:1712.03485

> <

INTRO – 802.11ad/ay STANDARDS FOR MMWAVE

8

802.11ad frame structure

Used to steer the beam towards the receiver

BTI A- BFT ATI SP/CBA P SP/CBA P SP/CBA P SP/CBA P

CDWN31 ID=14

BTI

The AP broadcasts beacon frames in a sectorized/brute force manner. Device replies back with preferred sector.

CDWN0 ID=4 CDWN=31 ID=15 BEST=5 CDWN=0 ID=4 BEST=5 CDWN=31 ID=3 BEST=1 CDWN=0 ID=9 BEST=1 CDWN=31 ID=13 BEST=8 CDWN=0 ID=2 BEST=8 Poll ID1 Poll ID4 SPR ID1

SECTOR LEVEL SWEEP (SLS)

SPR ID4

… … … …

AP polls to make an efficient use of the wireless spectrum.

ATI ATI – POLLING PHASE

… …

A-BFT

802.11ad MAC structure Very simplistic beamforming procedure that incurs in HIGH PROCESSING time. It does not exploit CSI or mmWave channel statistics.

> <

Aim of the project: Explore novel beamforming techniques aiming to accommodate more users in a highly demanding 5G environment.

[1] 10. Hong. W. “Study and prototyping of practically large-scale mmWave antenna systems for 5G cellular devices”. IEEE Commun. 2014 [2] Sadhu B. “A 28-GHz 32-Element TRX Phased-Array IC With Concurrent Dual-Polarized Operation and Orthogonal Phase and Gain Control for 5G Communications”. IEEE J. Solid-State Circuits. 2017 [3] Gu X. “A multilayer organic package with 64 dual-polarized antennas for 28GHz 5G communication”; Proceedings of the IEEE IMS2017. 2017

Minimize Inter-User Interference (IUI), maximizing the num. of users scheduled while keeping their PER at a minimum

OVERVIEW

9

Some numbers for mmWave deployments

• 16 antenna arrays for UEs [1].
• 64 [2,3], 128 (and more) antenna arrays for

BSs. [2] [3] [1]

> <

Contribution 1:

Cross-layer optimization (Network+Link+PHY) where antennas are allocated to users in time in order to meet their application requirements.

CROSS-LAYER APPROACH

10

INTERNET TRAFFIC MODELING (QoS per Application, arrivals, packet length, etc) SCHEDULING (Traffic aggregation and user scheduling) MMWAVE ARRAY (RF chain – antenna interconnection) MMWAVE CHANNEL (LoS vs NLOS, indoors vs outdoors) MU-MIMO BEAMFORMING (Dynamically formed sub-arrays) S Y S T E M

User 1 User 2

mmWave channel

Slot 0 User 1 User 2

INTERNET

Central controller - router Beamforming algorithm

> <

CROSS-LAYER APPROACH

11

INTERNET QoS per Application SCHEDULING (Traffic aggregation and user scheduling) MMWAVE ARRAY (RF chain – antenna interconnection) MMWAVE CHANNEL (LoS vs NLOS, indoors vs outdoors) MU-MIMO BEAMFORMING (Dynamically formed sub-arrays) S Y S T E M

Contribution 2:

Consider realistic deployment scenarios in a resource constrained system (limited number of antennas and RF). We explore non-uniform/unconventional antenna allocation and beamforming configurations for MU-MIMO ([1] only SU-SISO)

Conventional Unconventional

[1] S. Park, A. Alkhateeb and R. W. Heath, "Dynamic Subarrays for Hybrid Precoding in Wideband mmWave MIMO Systems," TWC, 2017.

User 1 User 2 User 3 User 4

> <

[1] A. Adhikary et al., "Joint Spatial Division and Multiplexing for mm-Wave Channels," in IEEE Journal on Selected Areas in Communications, 2014.

CROSS-LAYER APPROACH

12

INTERNET QoS per Application SCHEDULING (Traffic aggregation and user scheduling) MMWAVE ARRAY (RF chain – antenna interconnection) MMWAVE CHANNEL (LoS vs NLOS, indoors vs outdoors) MU-MIMO BEAMFORMING (Dynamically formed sub-arrays) S Y S T E M

Contribution 3:

We group users to be allocated per slot, not only accounting for the spatial diversity (like in [1]), but also considering asymmetric traffic and uneven demands.

…

User 1 User 2 User 3 User 4

QCI Priority

• Tol. Delay

PER Description 1 2 100 10−2

• Conv. Voice (Live)

2 4 150 10−3

• Conv. Video (Live)

3 3 50 10−3 Real Time Gaming 4 5 300 10−6 Non-Conv. Video 65 0.7 75 10−2 Mission Critical (Push2Talk) 66 2 100 10−2 Non-Mission Critical (Push2Talk) 5 1 100 10−6 IMS Signaling 6 6 300 10−6 Video (Live) 7 7 100 10−3 Video (Buffered) 8 8 300 10−6 Video (Buffered) 9 9 300 10−6 Video (Buffered) 69 0.5 60 10−6 Mission Critical Delay Sensitive 70 0.6 200 10−6 Mission Critical Data

time Slot 0 Slot 1 Slot 2 Slot N

> <

[1] V. Carela-Espanol, et al., “Is our ground-truth ˜ for traffic classification reliable?” in Passive and Active Measurement, M. Faloutsos and A. Kuzmanovic, Eds. Cham: Springer International Publishing, 2014, pp. 98–108. [2] T. Bujlow, et al., “Independent comparison of popular dpi tools for traffic classification,” Computer Networks, vol. 76, pp. 75 – 89, 2015.

CROSS-LAYER APPROACH

13

Contribution 4:

We accurately generate Internet traffic arrivals using real Internet public traces [1,2]. The traces capture over 700K flows, gathering more than 53 GB of raw data.

500 1000 1500

Bits Packet Arrivals in Bits for User 1

1 2 3 4 5 6 time (s) 104

Youtube

500 1000 1500

Bits Packet Arrivals in Bits for User 2

0.5 1 1.5 2 2.5 3 3.5 time (s) 105

Justin TV

5000 10000

Bits Packet Arrivals in Bits for User 3

1 2 3 4 5 6 time (s) 106

Facebook

500 1000 1500

Bits Packet Arrivals in Bits for User 4

2 4 6 8 10 12 14 time (s)

5

Web Browsing

INTERNET TRAFFIC MODELING (QoS per Application, arrivals, packet length, etc) SCHEDULING (Traffic aggregation and user scheduling) MMWAVE ARRAY (RF chain – antenna interconnection) MMWAVE CHANNEL (LoS vs NLOS, indoors vs outdoors) MU-MIMO BEAMFORMING (Dynamically formed sub-arrays) S Y S T E M

Properties:

• self-similarity
• heavy-tailed
• long-range

dependency

> <

14

INTERNET TRAFFIC - PARSER

• Num. of bits per Flow

Youtube Vimeo Justin TV Facebook Web Browsing Twitter Amazon SSH FTP 5 10 105

Flow Duration (s)

Youtube Vimeo Justin TV Facebook Web Browsing Twitter Amazon SSH FTP 20 40 60 80 100

• Num. of Flows

Youtube Vimeo Justin TV Facebook Web Browsing Twitter Amazon SSH FTP 1 2 3 104

• Num. of Packets per Flow

Youtube Vimeo Justin TV Facebook Web Browsing Twitter Amazon SSH FTP 500 1000 1500 2000

The distribution gives us a sense of application presence amongst users: Allows for a realistic traffic recreation.

Traffic distribution - Output

[1] V. Carela-Espanol, et al., “Is our ground-truth ˜ for traffic classification reliable?” in Passive and Active Measurement, M. Faloutsos and A. Kuzmanovic, Eds. Cham: Springer International Publishing, 2014, pp. 98–108. [2] T. Bujlow, et al., “Independent comparison of popular dpi tools for traffic classification,” Computer Networks, vol. 76, pp. 75 – 89, 2015.

𝝁 = Inter-packet arrival U1 time

Parsing the dataset in Matlab

𝑸 = Inter-packet arrival 𝑬 = Flow duration

• 53.31 GB of raw packet data
• 97.41% of it labeled with its application
• 750K Internet flows in total.

DATA SET Raw data PARSER TRAFFIC GENERATOR

𝝁 𝑸 𝑬

𝐄𝐁𝐔𝐁𝐓𝐅𝐔

𝒔𝒇𝒃𝒆 𝒇𝒓𝒗𝒋𝒘𝒃𝒎𝒇𝒐𝒖

> <

15

INTERNET TRAFFIC - SCHEDULER

Internet traffic generator (bits) Flow creator (Throughput) Compute for all nodes < 𝑼𝑰(𝒋)

(𝒖), 𝜺(𝒋) (𝒖) >

Compute Priorities 𝑸𝒔(𝒋)

(𝒖) =

𝟐 𝜺(𝒋)

(𝒖)

Create Sorted Combination list 𝜵(𝒖) = 𝜵(𝒖,𝟐), … , 𝜵 𝒖,𝒍 , …

𝜺(𝒋)

(𝒖)

𝑼𝑰(𝒋)

(𝒖𝟏)

Internet packet arrival : Delivery deadline

𝑼𝑰(𝒋)

(𝒖𝟐)

Time Slot

U1 U2 time time time

Work flow

…

time Slot 1 Slot N

𝑼𝑰(𝟐)

(𝟐) ; 𝜺(𝟐) (𝟐)

𝑼𝑰(𝟓)

(𝟐) ; 𝜺(𝟓) (𝟐)

𝑼𝑰(𝟐)

(𝑶) ; 𝜺(𝟐) (𝑶)

𝑼𝑰(𝟓)

(𝑶) ; 𝜺(𝟓) (𝑶)

INPUTS TO PHY-LAYER BEAMFORMER

* 𝜵 𝒖,𝒍 : List of users attempted to be scheduled at iteration ‘k’ at time ‘t’. * 𝑼𝑰(𝒋)

(𝒖): Required Throughput by user ‘i’.

time

> <

16

CROSS-LAYER IMPLICATIONS - EXAMPLE

PER of 10%

The Scheduler mandates the required throughput per user per slot, given the slot. The quality of the link depends on the performance of the Beamforming algorithm. The higher the Inter-User Interference (IUI), the lower the SINR and the higher the PER. A higher PER implies higher throughput demands to catch up with the tolerable application deadline. A lower PER lowers the retransmissions and eases the throughput requirements.

PER of 50%

> <

HELB: Heuristic Enhanced LCMV Beamformer

17

HELB – OUR BEAMFORMING ALGORITHM

Redistribute number of antennas (Genetic algorithm) and call LCMV Initialize Scoring function with LCMV on Random selection Determine number of antennas ∝ 𝑈𝐼(d)

(e)

Satisfy 𝑼𝑰 𝒖,𝒍 Efficient solution found for Ω e,g = h 𝑈𝐼(d)

(e) in

Polynomial time Re-evaluate user list subarray in Ω e,g with priorities yes no

Conventional beamforming Minimum Variance Distortionless Response (MVDR) Helps max. received power by steering the vector towards the direction of maximum (AoA). Linearly Constrained Minimum Variance (LCMV) Improvement from MVDR and allows for imposing nulls at reception to minimize interference (IUI). HELB exploits these features, LCMV does not:

• 1. Takes dissimilar SNR demands per user.
• 2. Exploits the flexibility of the antenna array.

BEAMFORMING WEIGHTS FOUND

> <

18

HELB – GENETIC ALGORITHMS (GA)

GA workflow:

Scoring function Crossover Mutation Genetic Algorithm Gene to Antenna assignation Initial Antenna assignation to Genes

Initial Assignation to Genes Suppose we have Na=6 antennas and Nu=3 users. We proceed as follow:

• 1. Convert the 2D array into a 1D array.
• 2. A gene could be g1 = [1 2 3 1 2 3].
• 3. User 1 gets antennas 1 and 4 (underlined)

Crossover HELB uses a single point crossover, for instance: g1 = [121233] and g2 = [232131]. A 50% crossover would lead to to g3 = [231231]. Mutation HELB uses a single pair order, for instance, the resulting gene g3, with for e.g., 33% mutation may result in g3 = [231213] GA configuration in HELB

> <

19

HELB – THE SCORING FUNCTION

The quality of the solution found in heuristics is measured by the following formula: Reward always between -1 and 1 Fitness_val= 𝜷 j 𝒕𝒅𝒑𝒔𝒇𝒋𝒖𝒆 + (𝟐 − 𝜷) j 𝒕𝒅𝒑𝒔𝒇𝒋𝒖𝒈

• 𝜷 controls the importance of achieving a higher directionality towards the intended user versus achieving a low directionality

to other users (between 0 and 1).

• 𝒕𝒅𝒑𝒔𝒇𝒋𝒖𝒆 and 𝒕𝒅𝒑𝒔𝒇𝒋𝒖𝒈 control the score based on the directivity achieved (dB) and the scoring function (between 0 and 1).
• 50dB is good enough.

Any extra suppression is not that meaningful The score takes negative values to penalize the poor behavior and force it to try other configurations

> <

20

HEURITSICS FOR BEAMFORMING – WHY?

Our system aims to answer the following 2 questions in each slot: (Q1) What users need to be allocated in each slot? Classical scheduling problem whose aim is to minimize the number of jobs scheduled after their completion deadline. The complexity is NP-Hard. (Q2) How should antennas, weights and phases be assigned (to users) and configured? Given the number of variables, the complexity of the problem is NP-Complete. Heuristics can be used on top of conventional Beamforming algorithms for ”result thinning”, achieving better performance.

Why using Heuristics?

The aim here is to maximize the Directivity towards the Intended user and minimize the interference towards any

• ther users.

Directivities to solve Q2

> <

21

THE MMWAVE CHANNEL

yi = WRX,iHi

M

X

u=1

FRF,uFBB,u + WRX,inu

HNLoS

u,s,n (t) =

r Pn M

M

X

m=1

FNLoS

u,n,mRu,n,mTs,n,mDn,m(t)

HLoS

u,s,n(t) =

r Pn M

M

X

m=1

FLoS

u,n,mRu,n,mTs,n,mDn,m(t)

PLoS = 8 > < > : 1, d2D ≤ 1.2m exp

• − d2D−1.2

4.7

• ,

1.2m < d2D < 6.5m exp

• − d2D−6.5

32.6

• ,

6.5m < d2D PLIn-LoS(dB) =17.3 log10 (d3D) + 20 log10 (fc) + 32.4 + ∆gLoS, 1m ≤ d3D ≤ 100m PL0

In-NLoS =38.3 log10(d3D) + 24.9 log10(fc)

+ 17.30 + ∆gNLoS, 1m ≤ d3D ≤ 86m PLIn-NLoS(dB) = max {PLIn-LoS, PL0

In-NLoS}

φn,m,AoA = φn,AoA + cASAαm

• 3. Probability of LoS/NLoS/Blockage
• 1. Path Loss
• 2. Angles of Arrival - Clustered
• 4. Channel Matrix
• The technical name is:

ETSI TR 138 901

• Good channel model for Physical

and Link layer simulations.

• Used in a wide variety of

environments (Indoors/Outdoors), similar to the TGn channel.

• Available at Mathworks central.

[1] 3GPP ETSI TR 38.901 , “5G, Study on channel model for frequencies from 0.5 to 100 GHz”, version 14.0.0, Release 14, 2017-05

Communication system model mmWave 3GPP channel model [1]

> <

22

PHY LAYER EMULATION

• Pkt. Detect &

Coarse Freq. Corr. Timing Sync & Channel Est. Extract Data Field & Reshape Bits required to be transmitted for 𝜵 𝒖,𝒍 Throughput ( h 𝑼𝑰(𝒋)

(𝒖)) required

for 𝜵 𝒖,𝒍 Guard Interval Insertion LDPC- Encoder Scrambler Modulation Pulse Shaping / Up Conversion MCS computation 5G mmWave channel

• Clustered MPC
• AoA/AoD
• Path Loss

802.11ad/ay frame mmWave link using Beamforming 802.11ad/ay TX block diagram Noise Est. Equalize Data Field Phase Tracking & Corr. DMG Data Recover

PACKET ERROR RATE (PER)

> <

23

MCS SELECTION IN 802.11ad/ay USING MATLAB

5 10 15 20

SNR (dB)

10-4 10-3 10-2 10-1 100

PER PER for DMG SC-PHY with 3GPP TR 38.901 Channel, CDL-C

5 10 15 20

SNR (dB)

10-4 10-3 10-2 10-1 100

PER PER for DMG SC-PHY with 3GPP TR 38.901 Channel, CDL-D

MCS 1 MCS 2 MCS 3 MCS 4 MCS 5 MCS 6 MCS 7 MCS 8 MCS 9 MCS 10 MCS 11 MCS 12

• App. Type 1
• App. Type 2
• App. Type 3

The PER limit is defined by the Table I The WLAN Toolbox allows us to create realistic PHY frames to send through the 3GPP channel created

• previously. Properties such as the size of the Payload are configured by default following the specs.

The MCS is selected following the standard procedure of the Look-up-Table, where the tolerable PER is defined by the application (previously detailed in Table I).

PER characterization for mmWave (Channels C and D)

[1] Mathworks example: https://www.mathworks.com/help/wlan/examples/802-11ad-packet-error-rate-simulation-for-ofdm-phy.html

> <

1

• 1
• 2
• 3

60 75 90 105 120 135 150 165 180 195 210 225 240 255 270 285 300 315 330 345 45 30 15

User 2 Phi: -11.25

24

RESULTS (1) - DIRECTIVITIES

1

• 1
• 2
• 3

60 75 90 105 120 135 150 165 180 195 210 225 240 255 270 285 300 315 330 345 45 30 15

User 1 Phi: -22.5

1

• 1
• 2
• 3

60 75 90 105 120 135 150 165 180 195 210 225 240 255 270 285 300 315 330 345 45 30 15

User 1 Phi: -22.5

• 1
• 2
• 3
• 4

60 75 90 105 120 135 150 165 180 195 210 225 240 255 270 285 300 315 330 345 45 30 15

User 2 Phi: -11.25

• 1
• 2
• 3
• 4

60 75 90 105 120 135 150 165 180 195 210 225 240 255 270 285 300 315 330 345 45 30 15

User 3 Phi: 11.25

• 1
• 2
• 3
• 4

60 75 90 105 120 135 150 165 180 195 210 225 240 255 270 285 300 315 330 345 45 30 15

User 3 Phi: 11.25

• 1
• 2
• 3
• 4
• 5

60 75 90 105 120 135 150 165 180 195 210 225 240 255 270 285 300 315 330 345 45 30 15

User 4 Phi: 22.5

• 1
• 2
• 3
• 4

60 75 90 105 120 135 150 165 180 195 210 225 240 255 270 285 300 315 330 345 45 30 15

User 4 Phi: 22.5

• 0.02
• 0.015
• 0.01
• 0.005

0.005 0.01 0.015 0.02

x

• 0.02
• 0.015
• 0.01
• 0.005

0.005 0.01 0.015 0.02

y Subarray Selection

• 0.02
• 0.015
• 0.01
• 0.005

0.005 0.01 0.015 0.02

x

• 0.02
• 0.015
• 0.01
• 0.005

0.005 0.01 0.015 0.02

y Subarray Selection

HELB returns an unconventional antenna allocation. HELB generates less interference in the intended direction at the expense of greater interference on other directions

LCMV HELB Directivities using a 16x16 (256) antenna array at the transmitter

> <

25

RESULTS (1) – CAPACITY AND SNR WITH HELB

PHY analysis: Proposed HELB vs LCMV

HELB’s performance is always as good as the fixed LCMV. HELB’s achieves a higher capacity as the number of users increases versus the LCMV. HELB’s offers the maximum aggregated capacity in the network for a higher number of users, increasing the users to be served together.

2 3 4 5 6 7 8 9 10 11 12

Number of users

• 80
• 60
• 40
• 20

20 40 60

SINR (dB) Average SINR achieved

144 antennas 196 antennas 256 antennas 324 antennas 400 antennas 484 antennas 2 3 4 5 6 7 8 9 10 11 12

Number of users

20 40 60 80 100 120 140

Capacity (bits/Hz/s) Total Capacity achieved

144 antennas 196 antennas 256 antennas 324 antennas 400 antennas 484 antennas 2 3 4 5 6 7 8 9 10 11 12

Number of users

2 4 6 8 10 12 14 16 18 20

Capacity (bits/Hz/s) Average Capacity achieved

144 antennas 196 antennas 256 antennas 324 antennas 400 antennas 484 antennas

> <

26

RESULTS (3) – CROSS-LAYER EVALUATION

1 2 3 4 5 6 7 8 9 10

Number of users

10 20 30 40 50 60 70 80 90 100

Ratio OK (%)

Ratio of data delivery - 64 antennas

Network sat. 0.08x Network sat. 0.21x Network sat. 0.25x Network sat. 0.33x Network sat. 0.42x Network sat. 0.54x Network sat. 0.67x Network sat. 1.00x

1 2 3 4 5 6 7 8 9 10

Number of users

10 20 30 40 50 60 70 80 90 100

Ratio OK (%)

Ratio of data delivery - 16 antennas

Network sat. 0.08x Network sat. 0.21x Network sat. 0.25x Network sat. 0.33x Network sat. 0.42x Network sat. 0.54x Network sat. 0.67x Network sat. 1.00x

Cross-layer analysis: Network saturation

Our simulator offers a complete Link-Phy layer simulation. The maximum network saturation (1.00x) matches the maximum capacity of the wireless channel The results depend upon (i) the antenna dimensions, (2) the number of users, (3) the channel profile (CDL-C,D, or

• ther), (4) the application presence amongs users, etc.

> <

27

MATLAB ToolBoxes

INTERNET TRAFFIC MODELING (QoS per Application, arrivals, packet length, etc) SCHEDULING (Traffic aggregation and user scheduling) MMWAVE ARRAY (RF chain – antenna interconnection) MMWAVE CHANNEL (LoS vs NLOS, indoors vs outdoors) MU-MIMO BEAMFORMING (Dynamically formed sub-arrays) S Y S T E M MATLAB Optimization toolbox MATLAB Communications toolbox MATLAB Phased Array Toolbox MATLAB WLAN Toolbox Ø Genetic Algorithms in HELB Ø Emulate the wireless channel Ø Generate 802.11ad standard’s compliant frames. Ø Compute Directivities in beamforming

1 2 3 4 1 2 3 4

> <

28

• 1. Proposed a cross-layer optimization (Network+Link+PHY) where antennas are allocated to

users in time (TDMA) as to guarantee the desired QoS mandated by application.

• 2. Considered realistic deployment scenarios in a resource constrained system (limited number
• f antennas and RF).
• 3. Explored non-uniform/unconventional antenna allocation and beamforming configurations

for MU-MIMO.

• 4. Applied a user grouping strategy in a user per slot basis, accounting for the spatial diversity

and heterogeneous traffic and uneven demands.

• 5. We validate our system using Internet traces to accurately emulate the arrivals and payload

distribution of a real deployment.

• 5. Github code: https://github.com/MathworksProjects/mmWave-MU-MIMO

Conclusions