On Self-interference Suppression Methods for Low-complexity Full-duplex MIMO

Transcription

1 On Self-interference Suppression Methods for Low-complexity Full-duplex MIMO Alexios Balatsoukas-Stimming, Pavle Belanovic, Konstantinos Alexandris, Andreas Burg Telecommunications Circuits Laboratory (TCL) École Polytechnique Fédérale de Lausanne (EPFL) Telecommunications Circuits Laboratory

2 Outline 1 Full-duplex architectures 2 Components of self-interference signal 3 Experimental results 4 Full-duplex MIMO 5 Conclusions 1/19

3 Introduction FD wireless promises improvements in maximum throughput spectral efficiency relaying medium access control 2/19

4 Introduction FD wireless promises improvements in maximum throughput spectral efficiency relaying medium access control The catch: strong self-interference 2/19

5 Introduction FD wireless promises improvements in maximum throughput spectral efficiency relaying medium access control The catch: strong self-interference Recently, several implementations shown that FD is feasible in practice 2/19

6 Analog construction of cancelation signal h tx : multipath self-interference channel ĥ tx : estimated and reproduced self-interference channel e tx : transmitter noise (phase noise, non-linearities,...) 3/19

7 Analog construction of cancelation signal Suppression β s E[(htx ĥtx)2 ] E[h 2 tx] Analog cancelation limited by how well h tx is reproduced h tx : multipath self-interference channel ĥ tx : estimated and reproduced self-interference channel e tx : transmitter noise (phase noise, non-linearities,...) 3/19

8 Analog construction of cancelation signal Suppression β s E[(htx ĥtx)2 ] E[h 2 tx] Analog cancelation limited by how well h tx is reproduced Residual z s = (s + e tx ) (h tx ĥtx) h tx : multipath self-interference channel ĥ tx : estimated and reproduced self-interference channel e tx : transmitter noise (phase noise, non-linearities,...) 3/19

9 Analog construction of cancelation signal Suppression β s E[(htx ĥtx)2 ] E[h 2 tx] Analog cancelation limited by how well h tx is reproduced Residual z s = (s + e tx ) (h tx ĥtx) + Residual interference can be treated in the digital domain 3/19

10 Analog construction of cancelation signal Suppression β s E[(htx ĥtx)2 ] E[h 2 tx] Analog cancelation limited by how well h tx is reproduced Residual z s = (s + e tx ) (h tx ĥtx) + Residual interference can be treated in the digital domain Relatively complex analog cancelation circuitry 3/19

11 Analog construction of cancelation signal Suppression β s E[(htx ĥtx)2 ] E[h 2 tx] Analog cancelation limited by how well h tx is reproduced Residual z s = (s + e tx ) (h tx ĥtx) + Residual interference can be treated in the digital domain Relatively complex analog cancelation circuitry Analog circuitry scales with N 2 in a MIMO system 3/19

12 Digital construction of cancelation signal h tx : multipath self-interference channel ĥ: cancelation channel e tx, e cx : transmitter and canceler noise, resp. 4/19

13 Digital construction of cancelation signal Suppression β r 2E[e2 tx] E[s 2 ] = 2 EVM Cancelation limited by transmitter noise h tx : multipath self-interference channel ĥ: cancelation channel e tx, e cx : transmitter and canceler noise, resp. 4/19

14 Digital construction of cancelation signal Suppression β r 2E[e2 tx] E[s 2 ] = 2 EVM Cancelation limited by transmitter noise Residual z r = s (h tx + }{{ ĥ hcx) + e tx h tx + e cx h }}{{ cx } channel estimate transmitter noises h tx : multipath self-interference channel ĥ: cancelation channel e tx, e cx : transmitter and canceler noise, resp. 4/19

15 Digital construction of cancelation signal Suppression β r 2E[e2 tx] E[s 2 ] = 2 EVM Cancelation limited by transmitter noise Residual z r = s (h tx + }{{ ĥ hcx) + e tx h tx + e cx h }}{{ cx } channel estimate transmitter noises Residual self-interference is difficult to suppress digitally 4/19

16 Digital construction of cancelation signal Suppression β r 2E[e2 tx] E[s 2 ] = 2 EVM Cancelation limited by transmitter noise Residual z r = s (h tx + }{{ ĥ hcx) + e tx h tx + e cx h }}{{ cx } channel estimate transmitter noises Residual self-interference is difficult to suppress digitally + Required analog circuitry is only another DAC and mixer 4/19

17 Digital construction of cancelation signal Suppression β r 2E[e2 tx] E[s 2 ] = 2 EVM Cancelation limited by transmitter noise Residual z r = s (h tx + }{{ ĥ hcx) + e tx h tx + e cx h }}{{ cx } channel estimate transmitter noises Residual self-interference is difficult to suppress digitally + Required analog circuitry is only another DAC and mixer Analog cancelation circuitry scales with N in a MIMO system 4/19

18 Suppression stages Passive analog Propagation loss Antenna separation, circulator, shielding, directional antennas,... 5/19

19 Suppression stages Passive analog Propagation loss Antenna separation, circulator, shielding, directional antennas,... Active analog Properly constructed cancelation signal is added in analog domain Suppression before the ADC 5/19

20 Suppression stages Passive analog Propagation loss Antenna separation, circulator, shielding, directional antennas,... Active analog Properly constructed cancelation signal is added in analog domain Suppression before the ADC Digital Residual interference estimated and removed in digital domain No effect if analog cancelation is good 5/19

21 Components of self-interference signal Dominant components: Line-of-sight interference (h tx s) 6/19

22 Components of self-interference signal Dominant components: Line-of-sight interference (h tx s) Multipath interference (h tx s) 6/19

23 Components of self-interference signal Dominant components: Line-of-sight interference (h tx s) Multipath interference (h tx s) Non-linear cancelation (h (1) tx s + h (2) tx s 2 + h (3) tx s ) 6/19

24 Components of self-interference signal Dominant components: Line-of-sight interference (h tx s) Multipath interference (h tx s) Non-linear cancelation (h (1) tx s + h (2) tx s 2 + h (3) tx s ) Phase noise (dominant component of EVM) 6/19

25 Components of self-interference signal Dominant components: Line-of-sight interference (h tx s) Multipath interference (h tx s) Non-linear cancelation (h (1) tx s + h (2) tx s 2 + h (3) tx s ) Phase noise (dominant component of EVM) Some energy may remain in the residual self-interference, z 6/19

26 Components of self-interference signal Dominant components: Line-of-sight interference (h tx s) Multipath interference (h tx s) Non-linear cancelation (h (1) tx s + h (2) tx s 2 + h (3) tx s ) Phase noise (dominant component of EVM) Some energy may remain in the residual self-interference, z How much of z is still cancelable in principle? 6/19

27 Finding the cancelable part of z Send the same data s repeatedly, K times z (k) = s (h (k) tx + ĥ h(k) cx ) + e (k) tx h(k) tx + e(k) cx h (k) cx, k {1, 2,..., K} 7/19

28 Finding the cancelable part of z Send the same data s repeatedly, K times z (k) = s (h (k) tx + ĥ h(k) cx ) + e (k) tx h(k) tx + e(k) cx h (k) cx, k {1, 2,..., K} Deterministic component (constant wrt data, cancelable) z = 1 K K k=1 z(k) Random component (variable, zero-mean) z (k) = z (k) z 7/19

29 Finding the cancelable part of z Send the same data s repeatedly, K times z (k) = s (h (k) tx + ĥ h(k) cx ) + e (k) tx h(k) tx + e(k) cx h (k) cx, k {1, 2,..., K} Deterministic component (constant wrt data, cancelable) z = 1 K K k=1 z(k) Random component (variable, zero-mean) z (k) = z (k) z Cancelability metric: ρ = 10 log 10 ( 1 K z 2 2 k z(k) 2 2 ) 7/19

30 Genie cancelation A genie would be able to predict the cancelable component 8/19

31 Genie cancelation A genie would be able to predict the cancelable component Take another Q measurements z (q), q {1, 2,..., Q} 8/19

32 Genie cancelation A genie would be able to predict the cancelable component Take another Q measurements z (q), q {1, 2,..., Q} Perform genie suppression z (q) = z (q) z 8/19

33 Genie cancelation A genie would be able to predict the cancelable component Take another Q measurements z (q), q {1, 2,..., Q} Perform genie suppression z (q) = z (q) z Obtain average additional suppression 10 log 10 ( 1 Q ) z (q) 2 2 q z (q) 2 2 8/19

34 Genie cancelation A genie would be able to predict the cancelable component Take another Q measurements z (q), q {1, 2,..., Q} Perform genie suppression z (q) = z (q) z Obtain average additional suppression 10 log 10 ( 1 Q ) z (q) 2 2 q z (q) 2 2 We observe that cancelable component z is uncorrelated with s k, k = 1, 2, 3,... data-dependent 8/19

35 Hardware setup National Instruments FlexRIO with 2x 5791R RF modules 2.4 GHz ISM band up to 8 dbm transmit power Measured noise 20 MHz is 85 dbm Transmitter noise is < 37 db (EVM) 9/19

36 Hardware setup National Instruments FlexRIO with 2x 5791R RF modules 2.4 GHz ISM band up to 8 dbm transmit power Measured noise 20 MHz is 85 dbm Transmitter noise is < 37 db (EVM) Two antenna front-ends 1. Circulator 2. Two antennas with 30 cm separation 9/19

37 Hardware setup National Instruments FlexRIO with 2x 5791R RF modules 2.4 GHz ISM band up to 8 dbm transmit power Measured noise 20 MHz is 85 dbm Transmitter noise is < 37 db (EVM) Two antenna front-ends 1. Circulator 2. Two antennas with 30 cm separation MATLAB interface for digital baseband processing (OFDM) and testbed configuration 9/19

38 Hardware setup Scenario 1: circulator 18 db passive suppression 10/19

39 Hardware setup Scenario 2: two antennas with 30 cm separation 39 db passive suppression 11/19

40 Cancelation results: circulator front-end 20 MHz BW, 4 dbm transmit power Powers(dBm) Txs(4dBm) NoisesFloors( 85dBm) Frequencys(MHz) 12/19

41 Cancelation results: circulator front-end 20 MHz BW, 4 dbm transmit power 20 0 Passive analog: -18 db 20 Poweri(dBm) Txi(4dBm) Rxi( 14dBm) NoiseiFloori( 85dBm) Frequencyi(MHz) 12/19

42 Cancelation results: circulator front-end 20 MHz BW, 4 dbm transmit power PowerpldBm) Passive analog: -18 db Active analog Linear: -37 db Non-linear: 0 db Txpl4dBm) Rxpl 14dBm) RFpSup.plsharedpref.)pl 51dBm) NoisepFloorpl 85dBm) FrequencyplMHz) 12/19

43 Cancelation results: circulator front-end 20 MHz BW, 4 dbm transmit power PowerpldBm) Passive analog: -18 db Active analog Linear: -37 db Non-linear: 0 db Red. phase jitter: -11 db 80 Txpl4dBm) 100 Rxpl 14dBm) RFpSup.plsharedpref.)pl 51dBm) NoisepFloorpl 85dBm) RFpSup.plsharedposc.)pl 62dBm) FrequencyplMHz) 12/19

44 Cancelation results: circulator front-end 20 MHz BW, 4 dbm transmit power PowerpgdBm Txpg4dBm3 Dig.pSup.pg 63dBm3 Rxpg 14dBm3 RFpSup.pgsharedpref.3pg 51dBm3 NoisepFloorpg 85dBm3 RFpSup.pgsharedposc.3pg 62dBm FrequencypgMHz3 Passive analog: -18 db Active analog Linear: -37 db Non-linear: 0 db Red. phase jitter: -11 db Digital: 0 db 12/19

45 Cancelation results: circulator front-end 20 MHz BW, 4 dbm transmit power PowerpgdBm Txpg4dBm3 Dig.pSup.pg 63dBm3 Rxpg 14dBm3 GeniepSup.pg 73dBm3 RFpSup.pgsharedpref.3pg 51dBm3 NoisepFloorpg 85dBm3 RFpSup.pgsharedposc.3pg 62dBm FrequencypgMHz3 Passive analog: -18 db Active analog Linear: -37 db Non-linear: 0 db Red. phase jitter: -11 db Digital: 0 db Genie: -11 db (ρ = 10 db) 12/19

46 Cancelation results: circulator front-end 20 MHz BW, 4 dbm transmit power PowerpgdBm Txpg4dBm3 Dig.pSup.pg 63dBm3 Rxpg 14dBm3 GeniepSup.pg 73dBm3 RFpSup.pgsharedpref.3pg 51dBm3 NoisepFloorpg 85dBm3 RFpSup.pgsharedposc.3pg 62dBm FrequencypgMHz3 Passive analog: -18 db Active analog Linear: -37 db Non-linear: 0 db Red. phase jitter: -11 db Digital: 0 db Genie: -11 db (ρ = 10 db) Total: -77 db 12/19

47 Cancelation results: circulator front-end 20 MHz BW, 4 dbm transmit power PowerpgdBm Txpg4dBm3 Rxpg 14dBm3 RFpSup.pgsharedpref.3pg 51dBm3 Dig.pSup.pg 63dBm3 GeniepSup.pg 73dBm3 NoisepFloorpg 85dBm3 RFpSup.pgsharedposc.3pg 62dBm FrequencypgMHz3 Residual power: -71 dbm Decent, but we need more! Passive analog: -18 db Active analog Linear: -37 db Non-linear: 0 db Red. phase jitter: -11 db Digital: 0 db Genie: -11 db (ρ = 10 db) Total: -77 db 12/19

48 Cancelation results: antenna front-end 20 MHz BW, 4 dbm transmit power Powers(dBm) Txs(4dBm) Noisesloors( 85dBm) Frequencys(MHz) 13/19

49 Cancelation results: antenna front-end 20 MHz BW, 4 dbm transmit power Passive analog: -39 db PowerN(dBm) TxN(4dBm) RxN( 35dBm) NoiseNloorN( 85dBm) FrequencyN(MHz) 13/19

50 Cancelation results: antenna front-end 20 MHz BW, 4 dbm transmit power 20 0 PowerSldBm) Passive analog: -39 db Active analog: -46 db TxSl4dBm) RxSl 35dBm) RFSSup.SlsharedSosc.)Sl 81dBm) NoiseSloorSl 85dBm) FrequencySlMHz) 13/19

51 Cancelation results: antenna front-end 20 MHz BW, 4 dbm transmit power 20 0 PowerSgdBmN Passive analog: -39 db Active analog: -46 db Digital: -2 db TxSg4dBmN RxSg 35dBmN RFSSup.SgsharedSosc.NSg 81dBmN Dig.SSup.Sg 83dBmN NoiseSloorSg 85dBmN FrequencySgMHzN 13/19

52 Cancelation results: antenna front-end 20 MHz BW, 4 dbm transmit power 20 0 PowerSgdBmN Passive analog: -39 db Active analog: -46 db Digital: -2 db Genie: 0 db (ρ = 6 db) 100 TxSg4dBmN RxSg 35dBmN RFSSup.SgsharedSosc.NSg 81dBmN Dig.SSup.Sg 83dBmN NoiseSloorSg 85dBmN FrequencySgMHzN 13/19

53 Cancelation results: antenna front-end 20 MHz BW, 4 dbm transmit power 20 PowerSgdBmN TxSg4dBmN RxSg 35dBmN RFSSup.SgsharedSosc.NSg 81dBmN Dig.SSup.Sg 83dBmN NoiseSloorSg 85dBmN FrequencySgMHzN Residual power: -83 dbm (at our noise floor) Passive analog: -39 db Active analog: -46 db Digital: -2 db Genie: 0 db (ρ = 6 db) Total: -87 db 13/19

54 Cancelation results: antenna front-end 20 MHz BW, 4 dbm transmit power 20 PowerSgdBmN TxSg4dBmN RxSg 35dBmN RFSSup.SgsharedSosc.NSg 81dBmN Dig.SSup.Sg 83dBmN NoiseSloorSg 85dBmN FrequencySgMHzN Residual power: -83 dbm (at our noise floor) -85 db suppression before the ADC! 13/19 Passive analog: -39 db Active analog: -46 db Digital: -2 db Genie: 0 db (ρ = 6 db) Total: -87 db

55 Cancelation results: antenna front-end Noise floor increase (db) MHz 20 MHz 40 MHz Tx power (dbm) Worst-case noise floor increase: 5 db (at 8 dbm transmit power) 14/19

56 2 2 FD-MIMO node 15/19

57 2 2 FD-MIMO node 15/19

58 2 2 FD-MIMO node Analog front-end scales like N 15/19

59 Self-interference in FD-MIMO r 1 = h 1,1 tx s 1 + h 2,1 tx s 2 r 2 = h 1,2 tx s 1 + h 2,2 tx s 2 Self-interference: through circulator ( -18 db) from second transmit antenna ( -20 db) 16/19

60 Self-interference in FD-MIMO r 1 = h 1,1 tx s 1 + h 2,1 tx s 2 r 2 = h 1,2 tx s 1 + h 2,2 tx s 2 Self-interference: through circulator ( -18 db) from second transmit antenna ( -20 db) Interference power is about 2 that of SISO setup Suppression for N-transmit antenna MIMO: N+1 N EVM 16/19

61 Self-interference in FD-MIMO r 1 = h 1,1 tx s 1 + h 2,1 tx s 2 r 2 = h 1,2 tx s 1 + h 2,2 tx s 2 Self-interference: through circulator ( -18 db) from second transmit antenna ( -20 db) Interference power is about 2 that of SISO setup Suppression for N-transmit antenna MIMO: N+1 N 4.5 db more suppression expected for 2 2 MIMO 3 db due to additional interference 1.5 db due to EVM scaling EVM 16/19

62 Performance 20 MHz BW, circulator, 20 cm TX1-TX2 antenna spacing Total active suppression [db] FD SISO FD MIMO Transmit power [dbm] Suppression results verify expectations 17/19

63 Conclusions After linear & non-linear cancelation and phase jitter reduction, genie reveals data dependent residual interference 85 db of cancelation before the ADC is possible More suppression is possible in MIMO due to EVM scaling Digital construction of cancelation signal provides favorable complexity scaling for MIMO 18/19

64 Thanks! Questions? 19/19