Dirty RF Impact on Interference Alignment

1

Dirty RF Impact on Interference Alignment

Per Zetterberg

2

Outline• Goal• Approach• Interference-alignment and CoMP• Implementation• Results• Impairment-modeling (closening the gap

theory-simulation)• Conclusion

3

Goal

TestbedMeasurements(USRP)

Impairment-modeling

FEREVMSINRMatch!

Detailedsimulation

New interesting and challenging techniquesAssumption: Results are general.

Channels

Robust approaches

4

Approach

Transmitter Spectrum analyzer

Test-bedMeasurements(USRP)

Impairment-modeling

Basic simulation

PC

Impairment model

5

Interference alignment

K-transmitters and K-receivers, K-links:K/2 simultaneous interference-free links.Requires coding over multiple channel realizations.Global channel knowledge required.

Cadambe/Jafar, ”Interference Alignment and Degrees of Freedom of the K-User Interference Channel”,IEEE Trans, Information Theory 2008.

6

Interference-alignment incarnationsIn frequency-domain:

Something new –will be studied later.

In antenna-domain:Co-ordinated beam-forming.

7

Co-ordinated Multi-Point CoMP

8

Implementation IA

BS 1

BS 2

BS 3

MS 1

MS 2

MS 3

Feedback:Wired ethernet

𝒗 1

𝒗 2

𝒗 3

𝒖1

𝒖2

𝒖3

9

Implementation: CoMP

BS 1

BS 2

BS 3

MS 1

MS 2

MS 3

Feedback:Wired ethernet

𝒗 1

𝒗 2

𝒗 3

𝒖1

𝒖2

𝒖3

10

Beamformer

SNIR𝑘=𝒖𝑘

∗𝑯 𝑘 ,𝑘𝒗 𝑘𝒗𝑘∗𝑯 𝑘 ,𝑘

∗ 𝒖𝑘

∑𝑛 ≠𝑘

𝒖𝑘∗𝑯 𝑘 ,𝑛 𝒗𝑛𝒗𝑛

∗𝑯 𝑘,𝑛∗ 𝒖𝑘

=¿

“Approaching the Capacity of Wireless Networks through Distributed Interference Alignment", by Krishna Gomadam, Viveck R. Cadambe and Syed A. Jafar.

Formulate virtual uplink SINR. Iterate

11

Frames

Payload10 OFDM symbols

Payload10 OFDM symbols

CSI referencesignals

Demodulation reference signals

38 subcarriers, 312.5kHz carrier-spacingQPSK, …., 256QAM0.25, 0.5, 0.75 –rate LDPC codes

• MS feed-back CSI to BS1.• BS1 calculate beam-formers.• BS1 sends weights to BS2,

BS3.• BS1-BS3 frequency locked.

12

Preliminary Results

16QAM, 0.75 rate coded. 432 frames transmitted.

FER IA CoMPUncoded 63% 9%Coded 19% 0%

13

How far from ideal ?

0 5 10 15 20 25 30 35-5

0

5

10

15

20

25

30

35

Receiver estimated SNR

Act

ual S

NR

IA

0 5 10 15 20 25 30 350

5

10

15

20

25

30

35

Receiver estimated SNR

Act

ual S

NR

CoMP

d

e

PP100EVM%

14

Power-Amplifier Non-linearityOFDM signals:

+𝑠 (𝑡 )

n

y+n(t)

Modeled as noise:D Dardari, V. Tralli, A Vaccari “A theoretical characterization of nonlinear distortion effects in OFDM systems“, IEEE Trans. Comm., Oct 2000.

15

MIMO case

OFDM signals:

+𝑠1 (𝑡 )

n

(t)

OFDM signals:

+

n

(t)

𝑠2 (𝑡 )

Correlation ?

16

Phase-noise

A/D

ttfjt RX2expLO

LPFBPF LNA

y(t)

Modeled as additive noise + CPE

CPE: Slowly varying between symbols

R. Corvaja, E. Costa, and S. Pupolin, “M-QAM-OFDM system performance in the presence of a nonlinear amplifier and phase noise, IEEE Trans. Comm. 2002.

17

CoMP Results

0 5 10 15 20 25 30 350

10

20

30

40

0 5 10 15 20 25 30 350

10

20

30

40

Withoutimpairmentmodel

Withimpairmentmodel

18

IA Results

0 5 10 15 20 25 30 35-10

0

10

20

30

40

0 5 10 15 20 25 30 35-10

0

10

20

30

40

Withoutimpairmentmodel

Withimpairmentmodel

19

Conclusion: What will be answered?• How much worse is IA practice than in theory?• What practical impairments need to be modeled?

(==> can lead to improved robust designs)• Is IA still worthwhile with impairments compared to

base-lines ?

Other outputs• Software environments that can be re-used

(commodity hardware)• Increased understanding of software and hardware

issues and implementations in our research community and our PhDs in particular.

• Course-work for the above.

20

Structure of model

TX TX

RX TX

RX

RX

Channel

TX- impairment

21

Our Hardware

PC

Linux

Gbit-Ethernet

USRP N210Sample-rate: 100MHzStreaming: 25MHz

We have 18 USRPs

GPS

PPS,10MHz

22

The 4Multi Software FrameWork(Multi-Antenna, Multi-User, Multi-Cell, Multi-Band)

• Send data in small bursts (relaxes computational load)• Nodes synchronized by external trigering (PPS)• The implementor (basically) only need to program three functions node::init, node::process and node::end_of_run.• Simulate the system using “simulate” generic function.• Everything that can be compiled with gcc can run (e.g IT++)• Toolbox with coding&modulation.• Store _all_ received signals for post-processing.

Vision: “The coding should be as easy as performing ordinary

(but detailed) desktop simulations”