MIDU: Enabling MIMO Full Duplex Ehsan Aryafar Amir Khojastepour - PowerPoint PPT Presentation

MIDU: Enabling MIMO Full Duplex Ehsan Aryafar Amir Khojastepour Karthik Sundaresan Princeton NEC Labs NEC Labs Sampath Rangarajan Mung Chiang NEC Labs Princeton ACM MobiCom 2012 ACM MobiCom 2012

Background Background AP AP p 2 p 1 • Current wireless radios are half duplex C t i l di h lf d l Amir Khojastepour NEC Laboratories America

Background Background AP AP p 2 p 1 • Current wireless radios are half duplex C t i l di h lf d l • Same band Full duplex is hard – Self interference is very high: ≈ 75 dB for 15 dBm Tx power Self interference is very high 75 dB for 15 dBm Tx power – Transmitted signal is known  self interference cancellation – Self interference can be significantly reduced by adding a g y y g cancellation circuit: e.g., a cancelling antenna Amir Khojastepour NEC Laboratories America

Background Background AP AP p 2 p 1 • Current wireless radios are half duplex C t i l di h lf d l • Same band Full duplex is hard – Self interference is very high: ≈ 75 dB for 15 dBm Tx power Self interference is very high 75 dB for 15 dBm Tx power – Self interference can be significantly cancelled by adding a cancellation circuit: e.g., a cancelling antenna Can full duplex wireless double the capacity? Amir Khojastepour NEC Laboratories America

Full Duplex vs. MIMO Full Duplex vs. MIMO • Hardware complexity, performance, size, cost metrics Amir Khojastepour NEC Laboratories America

Full Duplex vs. MIMO Full Duplex vs. MIMO • Hardware complexity, performance, size, cost metrics • • Antenna Conserved (AC): Same Antenna Conserved (AC): Same # antennas Amir Khojastepour NEC Laboratories America

Full Duplex vs. MIMO Full Duplex vs. MIMO • Hardware complexity, performance, size, cost metrics • • Antenna Conserved (AC): Same Antenna Conserved (AC): Same # antennas • RF ‐ Chain Conserved (RC): Same # chains Amir Khojastepour NEC Laboratories America

Full Duplex vs. MIMO Full Duplex vs. MIMO • Hardware complexity, SI loss: 6 dB Ant Correlation: 0.1 performance, size, cost metrics  = 0.01  • • Antenna Conserved (AC): Same Antenna Conserved (AC): Same 100 FD-RC # antennas nel use) HD 80 • FD-AC RF ‐ Chain Conserved (RC): Same ty (bit/chann # chains 60 40 • Significant FD gains in RC model Significant FD gains in RC model Capaci • 20 Limited FD gains with small # antennas in AC model higher 0 0 0 5 5 10 10 15 15 20 20 gains with more antennas gains with more antennas Number of Antennas Amir Khojastepour NEC Laboratories America

Full Duplex vs. MIMO Full Duplex vs. MIMO • Hardware complexity, SI loss: 6 dB Ant Correlation: 0.1 performance, cost metrics  = 0.01  • • Antenna Conserved (AC): Same Antenna Conserved (AC): Same 100 FD-RC # antennas Regions of pronounced full duplex gains in nel use) HD 80 • FD-AC RF ‐ Chain Conserved (RC): Same b th RC both RC and AC models d AC m d ls ty (bit/chann # chains 60 40 • Significant FD gains in RC model Significant FD gains in RC model Capaci • 20 Limited FD gains with small # antennas in AC model high 0 0 0 5 5 10 10 15 15 20 20 gains with more antennas gains with more antennas Number of Antennas Amir Khojastepour NEC Laboratories America

Outline Outline • Background Background • Design of MIDU i f • Experimental Evaluation • Conclusion Amir Khojastepour NEC Laboratories America

MIDU: MImo full DUplex MIDU: MImo full ‐ DUplex • Symmetric antenna R1 T1 T'1 1 1 1 placement d d d d π Input Signal Input Signal Amir Khojastepour NEC Laboratories America

MIDU: MImo full ‐ DUplex MIDU: MImo full DUplex • Symmetric antenna RX Chain placement R1 • 2 level of antenna • 2 ‐ level of antenna cancellation T'1 T1 – Tx cancellation followed by π Rx cancellation Rx cancellation – Proved in theory to have additive gains under R'1 imbalanced gains/phase or imbalanced gains/phase or imprecise placement π TX Chain Amir Khojastepour

MIDU: MImo full ‐ DUplex MIDU: MImo full DUplex • Symmetric antenna placement R3 R2 • 2 level of antenna • 2 ‐ level of antenna R1 cancellation – Tx cancellation followed by Rx cancellation Rx cancellation – Proved in theory to have T'1 T'2 T'3 T1 T2 T3 additive gains under imbalanced gains/phase or imbalanced gains/phase or R'1 imprecise placement R'2 • Easy scalability to MIMO l b l R'3 Amir Khojastepour NEC Laboratories America

Implementation Implementation • WarpLab implementation – Narrow ‐ band 625 KHz – Open space environment – MIDU + MU ‐ MIMO MIDU + MU MIMO Virtex ‐ IV Pro FPGA Amir Khojastepour NEC Laboratories America

Implementation Implementation • WarpLab implementation – Narrow ‐ band 625 KHz – Open space environment – MIDU + MU ‐ MIMO MIDU + MU MIMO • Performance metric: SNR or the corresponding Shannon capacity di Sh i Virtex ‐ IV Pro FPGA Amir Khojastepour NEC Laboratories America

Implementation Implementation • WarpLab implementation – Narrow ‐ band 625 KHz – Open space environment – MIDU + MU ‐ MIMO MIDU + MU MIMO • Performance metric: SNR or the corresponding Shannon capacity di Sh i Virtex ‐ IV Pro FPGA • Spectrum analyzer based • Spectrum analyzer based measurement or the SNR reported by WARP Amir Khojastepour NEC Laboratories America

Experimental Evaluation Experimental Evaluation • Feasibility – Channel–distance relationship R3 – Stability R2 – Impact on far ‐ field users Impact on far field users R1 • Cancellation – Single ‐ level T'1 T'2 T'3 T1 T2 T3 – 2 ‐ level and MIMO • Comparison with MIMO R'1 – Single link R'2 – Single cell Single cell R'3 Amir Khojastepour NEC Laboratories America

Impact of MIDU on Far ‐ Field Users ld • Issue: How does symmetric y antenna placement impact the far ‐ field users? Amir Khojastepour

Impact of MIDU on Far ‐ Field Users ld 13 14 • Issue: How does symmetric y antenna placement impact the far ‐ field users? 12 15 3 4 2 5 16 11 TX 6 1 7 10 8 9 17 20 20 <3m> 19 18 Amir Khojastepour

Impact of MIDU on Far ‐ Field Users ld • Issue: How does symmetric y antenna placement impact the far ‐ field users? • Achieved SNR can be up to 4 dB higher/lower Outer Circle Inner Circle Amir Khojastepour

Impact of MIDU on Far ‐ Field Users ld • Issue: How does symmetric y antenna placement impact the far ‐ field users? • Achieved SNR can be up to 4 dB higher/lower • In far ‐ field antenna I f fi ld t cancellation has very limited effect due to signal scattering (fading) g ( g) • Similar results hold for RX cancellation Outer Circle Inner Circle Amir Khojastepour

Experimental Evaluation Experimental Evaluation • Feasibility – Channel–distance relationship R3 – Stability R2 – Impact on far ‐ field users Impact on far field users R1 • Cancellation – Single ‐ level T'1 T'2 T'3 T1 T2 T3 – 2 ‐ level and MIMO • Comparison with MIMO R'1 – Single link R'2 – Single cell Single cell R'3 Amir Khojastepour NEC Laboratories America

Cancellation Cancellation R1 T1 • T'1 Issue: Is 2 ‐ level cancellation 1 1 1 additive? Is MIDU scalable? • • Connect the receiver to a Connect the receiver to a d d d d spectrum analyzer π Input Signal Input Signal Amir Khojastepour NEC Laboratories America

Cancellation Cancellation • Issue: Is 2 ‐ level cancellation additive? Is MIDU scalable? • • 22 22 – 30 dB cancellation on each 30 dB cancellation on each level separately • Cancellation remains relatively unchanged with Tx power Amir Khojastepour NEC Laboratories America

Cancellation Cancellation RX Chain • Issue: Is 2 ‐ level cancellation additive? Is MIDU scalable? R1 Θ • • Phase shifter on each path to Phase shifter on each path to Θ + π T'1 T1 handle insertion loss and delay R'1 Θ Θ + π TX Chain Amir Khojastepour NEC Laboratories America

Cancellation Cancellation • Issue: Is 2 ‐ level cancellation additive? Is MIDU scalable? • • Phase shifter on each path to Phase shifter on each path to handle insertion loss and delay • RX cancellation on top of TX cancellation is additive Amir Khojastepour NEC Laboratories America

Cancellation Cancellation • Issue: Is 2 ‐ level cancellation additive? Is MIDU scalable? • • Phase shifter on each path to Phase shifter on each path to handle insertion loss and delay • RX cancellation on top of TX cancellation is additive • 4 dB decrease in cancellation for the first added pair , 5 dB with 3 total pairs with 3 total pairs Amir Khojastepour NEC Laboratories America

Experimental Evaluation Experimental Evaluation • Feasibility – Channel–distance relationship R3 – Stability R2 – Impact on far ‐ field users Impact on far field users R1 • Cancellation – Single ‐ level T'1 T'2 T'3 T1 T2 T3 – 2 ‐ level and MIMO • Comparison with MIMO R'1 – Single link R'2 – Single cell Single cell R'3 Amir Khojastepour NEC Laboratories America