Modulation using Smart(er) Antennas for 5G

Transcription

1 Modulation using Smart(er) Antennas for 5G A. Chockalingam Department of ECE Indian Institute of Science, Bangalore ECE Faculty Colloquium 28 July 217 (Joint work with Y. Naresh, Bharath Shamasundar, Swaroop Jacob)

2 Outline

3 Wireless spectrum Source: Internet

4 Modulation approaches Conventional view Symbols from complex alphabet (e.g., QAM/PSK) convey information bits Quadrature Amplitude In-phase Amplitude Modulation alphabet A

5 Modulation approaches Conventional view Channel fades viewed as causing amplitude/phase distortion to transmitted symbols Quadrature Amplitude In-phase Amplitude h 1x, x 2 A Quadrature Amplitude In-phase Amplitude h 2x, x 2 A h 1 ; h 2 2 CN (; 1): Complex channel fades Detrimental to performance

6 Modulation approaches Alternate view View complex channel fade coecients themselves to constitute a alphabet H = fh i ; i = 1; ; Mg as the alphabet Quadrature Amplitude In-phase Amplitude Channel alphabet H How to create this alphabet? transmit antennas, RF mirrors (parasitic elements)

7 Space shift keying Tone RF chain and 1 n t RF switch 1 2 n t h 1 1 h 2 2 hn t nr Tx Rx # tx. antennas, n t > 1; # tx. RF chains, n rf = 1 Constellation: H ssk = fh 1; h 2; ; h nt g Y. A. Chau and S.-H. Yu, \Space on wireless fading channels," in Proc. IEEE 54th VTC'21 (Fall), vol. 3, Oct. 21, pp J. Jeganathan, A. Ghrayeb, L. Szczecinski, and A. Ceron, \Space shift keying for channels," IEEE Trans. Wireless Commun., vol. 8, no. 7, pp , Jul. 29.

8 Parasitic elements Parasitic elements capacitors, varactors or switched capacitors that can adjust the resonance frequency Use of parasitic elements external to antennas Applications beamforming, DoA estimation selection/switched diversity recongurable antennas Indexing using parasitic elements aerial : index orthogonal antenna patterns realized using parasitic elements media-based : index channel fades realized using RF mirrors O. N. Alrabadi, A. Kalis, C. B. Papadias, R. Prasad, \Aerial for high order PSK transmission schemes," in Wireless VITAE 29, May 29, pp R. Bains, \On the usage of parasitic antenna elements in wireless communication," Ph.D. dissertation, Dept. Electron. Telecommun., Norwegian Univ. Sci. Technol., Trondheim, Norway, May 28. A. K. Khandani, \ : A new approach to wireless transmission," in Proc. IEEE ISIT'213, Jul. 213, pp

9 M rfmirrors MBM-TU Tone Tx RF chain ON/OFF control (m rf bits) m rf M rf # tx. antennas, n t = 1; # tx. RF chains, n rf = 1 # RF mirrors available, M rf ; # RF mirrors used, m rf Mirrors act as digitally controlled scatterers ON/OFF status of mirrors (a.k.a `Mirror Activation Pattern') create independent fade realizations Constellation: H mbm = fh 1 ; h 2 ; ; h 2 m rf g [A] A. K. Khandani, \ : A new approach to wireless transmission," in Proc. IEEE ISIT'213, Jul. 213, pp [B] A. K. Khandani, \ : Converting static Rayleigh fading to AWGN," in Proc. IEEE ISIT'214, Jun-Jul. 214, pp [C] E. Sei, M. Atamanesh, and A. K. Khandani, \ : A new frontier in wireless communications," online: arxiv: v3 [cs.it] 7 Oct 215. Rx 1 2 nr

10 SSK, MBM SSK Multiple tx. antennas create channel fade alphabet H ssk Advantage simple, need only 1 tx. RF chain, no interference Issue jh ssk j = n t, ssk = blog 2 n t c bpcu need exponential increase in n t to increase bpcu

11 SSK, MBM SSK MBM Multiple tx. antennas create channel fade alphabet H ssk Advantage Issue simple, need only 1 tx. RF chain, no interference jh ssk j = n t, ssk = blog 2 n t c bpcu need exponential increase in n t to increase bpcu Multiple RF mirrors create channel fade alphabet H mbm Advantage jh mbm j = 2 m rf ; mbm = m rf bpcu Issue bpcu increases linearly with m rf need to estimate jh mbm j = 2 m rf constellation points at the rx. through pilot transmission

12 MBM capacity MBM with 1 tx. antenna and n r rx. antennas achieves the capacity of n r parallel AWGN channels as m rf! 1. C in bits/s/hz nr = 2 - MBM nr =2 - parallel AWGN nr = 3 - MBM nr =3 - parallel AWGN nr = 4 - MBM nr =4 - parallel AWGN SNR = 2 db m rf

13 MBM implementation Source: E. Sei, M. Atamanesh, and A. K. Khandani, \ Modulation: Improving Spectral Eciency Beyond Conventional," E&CE Department, University of Waterloo.

14 MBM implementation.82mm Gaps are filled with a combination of metal, air and/or RF absorbers to have a good S11 for all patterns, without leaving too much energy out of the cavity prior to being affected by the switching pattern. PIN Diode Antenna.65mm Air Filled Opening Antenna 46.75mm 5.75mm 7.25mm Structure of each wall (RF mirror). 12.5mm Exterior metallic strips are placed around an external cylinder with openings to form a cavity. Reflections between walls of the external cylinder enriches the channel variations caused by switching of RF ON/OFF mirrors (walls of the interior cylinder). 11.2mm 26.2mm (c) (d) RF mirror (e) Cylindrical structure with RF mirrors Antenna patterns Source:

15 indoor/outdoor environments Indoor (residential building with dry-walls) Q Outdoor Model (down-town Ottawa) Q I I Indoor Outdoor MBM constellation points Source:

16 MBM: An instance of index Index bits are conveyed through indices of transmit entities Examples Indexing in spatial domain in multiantenna SSK, SM, GSM Indexing in frequency domain in multicarrier subcarrier index in OFDM Indexing in space and frequency GSFIM (generalized space-frequency index ) Indexing in space and time STIM (space-time index ) Indexing precoders PIM (precoder index ) Indexing RF mirrors MBM (media-based )

17 MBM signal set MBM + conventional A: a conventional alphabet (e.g., QAM) Dene A, A [ and N m, 2 m rf MBM signal set: set of N m 1-sized vectors given by S mbm = s 2 A Nm : s = [ s ] T ; {z} kth coordinate s 2 A; k = 1; ; N m where k is the index of the MAP. js mbm j = N m jaj Example: for m rf = 2 and jaj = 2 (i.e., BPSK), >< S mbm = ; 6 4 >: ; ; ; ; 6 4 MBM signal vectors are sparse vectors n r 1 rx. signal vector: y = sh k + n; ; ; >= 7 5 >; k: MAP index

18 MBM performance Bit error rate 1-1 SM: nt = 4,nrf = 1, 64-QAM (Spat. Mux.): nt = nrf = 2, 16-QAM GSM: nt = 4,nrf = 2, 8-QAM (Spat. Mux.): nt = nrf = 4, 4-QAM GSM: nt = 4,nrf = 3, 4-QAM -MBM: ntu = nrf = 2,mrf = 2, 4-QAM SIMO-MBM: ntu = nrf = 1,mrf = 6, 4-QAM 8 bpcu, MLD nr = Average SNR in db Figure : Comparison between SIMO-MBM with RF mirrors and other multi-antenna schemes without RF mirrors (, SM, GSM). MBM achieves very good performance

19

20 transmitter log 2 ( ntu nrf) bits log 2 M bits log 2 M bits MBM-TU Activation pattern selector QAM/PSK mapper QAM/PSK mapper m rf bits RF mirror index bits m rf bits RF chain RF chain n rf n tu RF mirrors n rf 1 n rf control switch n rf n tu RF switch n tu 1 2 n tu MBM-TU 1 MBM-TU 2 MBM-TU ntu Y. Naresh and A. Chockalingam, \On media-based using RF mirrors," in Proc. ITA'216, San Diego, Feb Also in IEEE Trans. Veh. Tech., vol. 66, no. 6, pp , Jun. 217.

21 Information bits are conveyed through MBM-TU indexing n rf out of n tu MBM-TUs selected using blog 2 n tu n rf c bits M-ary (QAM/PSK) symbols n rf M-ary symbols (formed using n rf log 2 M bits) are sent on the selected MBM-TUs RF mirror indexing ON/OFF status of m rf mirrors (mirror activation pattern) conveys m rf bits per MBM-TU Achieved rate in = j log 2 n tu! k n rf {z } MBM-TU index bits + n rf m rf + n rf log {z } 2 M bpcu {z } mirror index bits QAM/PSK symbol bits

22 system model # MAPs per MBM-TU: N m = 2 m rf S m : Set of all N m MAPs per MBM-TU H j = fh j 1 ; hj 2 ; ; hj N m g: MBM alphabet of jth MBM-TU h j k = [hj 1;k hj 2;k hj n r ;k ]T h j i ;k CN (; 1): channel fade corresponding to the kth MAP of jth MBM-TU to the ith receive antenna Received signal vector y = Xn tu j=1 s j h j l j + n; s j 2 A l j 2 f1; ; N m g: index of MAP chosen on jth MBM-TU y = Xn tu j=1 s j H j e lj + n; H j = [h j 1 hj 2 hj N m ]

23 system model Received signal vector can be written in the form y = Hx + n; H = [H 1 H 2 H ntu ] x belongs to the signal set given by n x = [x T 1 xt 2 x T n tu ] T : x j = s j e lj ; l j 2 f1; ; N m g; s = [s 1 s 2 s ntu ] T 2 S gsm o S gsm-mbm = ML decision rule is ^x = argmin ky Hxk 2 x2s gsm-mbm

24 performance Bit error rate 1 SIMO-MBM: ntu = nrf = 1,mrf = 4, 64-QAM (Sim.) SIMO-MBM: ntu = nrf = 1,mrf = 4, 64-QAM (Ana.) -MBM: ntu = nrf = 2,mrf = 2, 8-QAM (Sim.) 1-1 -MBM: ntu = nrf = 2,mrf = 2, 8-QAM (Ana.) : ntu = 4,nrf = 2,mrf = 2, 4-QAM (Sim.) : ntu = 4,nrf = 2,mrf = 2, 4-QAM (Ana.) MBM SIMO-MBM 1 bpcu, MLD nr = Average SNR in db Figure : Performance of SIMO-MBM, -MBM, and with n r = 8, and 1 bpcu.

25 MAP selection An MBM-TU may be designed to have more RF mirrors than actually used (M rf m rf ) 2 m rf best MAPs among 2 M rf MAPs can be selected (similar to tx. antenna selection in multiantenna ) Consider MAP selection in -MBM S all : set of all possible MAPs per MBM-TU. js all j = 2 M rf S sub : possible subset of S all. js sub j = 2 m rf Rx. estimates all js all j constellation points, selects the best js sub j among them, and sends the indices of the corresponding MAPs to the Tx. Tx. uses these selected MAPs to index the mirrors

26 Energy-based MAP selection Choose the MBM constellation points with the highest energies L j MI = fl j1 ; l j2 ; ; l jjs sub j g: set of MAP indices corresponding to the js sub j largest energies for the jth MBM-TU kh j l j1 k 2 kh j l j2 k 2 kh j l j js sub j k 2 kh j l j js all j k 2 I(h) ntu = 1,Mrf = 5,nr = 1,dmin = R(h) (a) All constellation points I(h) ntu = 1,Mrf = 5,mrf = 3 nr = 1,dmin = R(h) (b) Energy-based selection

27 ED-based MAP selection I j : collection of sets of MAP indices corresponding to the enumerations of the js all j combinations of selecting js sub j js sub j out of js all j MAPs of the jth MBM-TU L = L = fl 1 ; L 2 ; ; L ntu g : L j 2 I j ; j = f1; ; n tu Choose the set from L such that L ED = argmax L2L min x1 ;x 22X x16=x2 jjh L (x 1 x 2 ) jj 2 ntu = 1,Mrf = 5,nr = 1,dmin = ntu = 1,Mrf = 5,mrf = 3, BPSK nr = 1, dmin = I(h) I(h) R(h) R(h) (c) All constellation points (d) ED-based selection

28 MAP selection performance Bit error rate 1 Without MAP selection (Mrf = mrf = 1) MI-based MAP selection (Mrf = 2,mrf = 1) ED-based MAP selection (Mrf = 2,mrf = 1) -MBM: ntu = 2,nrf = 2 BPSK, 4 bpcu, MLD nr = Average SNR in db Figure : Performance of -MBM schemes without and with MAP selection. n tu = n rf = 2, BPSK, 4 bpcu, n r = 2.

29 MAP selection performance Diversity order of ED-based MAP selection: d = n r js all j js sub j ntu = 2,nrf = 2,mrf = 1 4 bpcu, MLD Bit error rate 1-3 d = Mrf = 2, BPSK, nr = 1 c1/snr 3 Mrf = 2, BPSK, nr = 2 d = c2/snr Average SNR in db

30

31 Conventional system K : # uplink users (in 1's) Each user has n t = 1 tx. antenna uses conventional n r : # rx. antennas at BS (in 1's)

32 MU-MBM system User 2 RF mirror ON/OFF control Tx Baseband Tx RF chain User 1 RF mirror ON/OFF control Tx Baseband Tx RF chain User K RF mirror ON/OFF control Tx Baseband Tx RF chain m rf RF mirrors 1 2 n r Base Station (1s to 1s of Rx antennas) Each user has n t = 1 tx. antenna and m rf uses MBM for signal transmission RF mirrors MBM signal set: S mu-mbm = S K mbm B. Shamasundar, A. Chockalingam, \ Modulation for Massive Systems," in Proc. IEEE SPAWC'217, Sapporo, Japan, Jul. 217.

33 MU-MBM rx. signal MU-MBM rx. signal: y = Hx + n y 1 y 2 y nr n r 1 Rx signal h 1 1 h2 1 hnm 1 h 1 2 h2 2 hnm 2 h 1 K h2 K hnm K H = [H 1H 2 H K] nr KNm K- user MBM channel User 1 MBM signal, x 1 s 1 s 2 n 1 n 2 User K MBM signal,x K s K KN m 1 K-user Tx signal n nr n r 1 Noise CN(,σ 2 I)

34 Sparsity-exploiting detection MBM signal vectors are inherently sparse An MBM signal vector has only one non-zero element out of N m elements, leading to a sparsity factor of 1=N m Example: Let K = 16, m rf = 4; N m = 2 m rf = 16. Out of KN m = 256 elements, only 16 are non-zeros, resulting in a sparsity factor of = 1 16 This inherent sparsity can be exploited for low complexity detection Sparse recovery algorithms like OMP, CoSaMP, and subspace pursuit (with necessary modications) can be used

35 Sparsity-exploiting detection Algorithm : Sparsity-exploiting detection of MU-MBM signals Inputs: y; H; K Initialize: j = repeat ^x r = SR(y; H; K + j). Sparse Recovery algorithm u j = UAP(^x r ). Extract User Activity Pattern if ku j k = K for q = 1 to K ^x q = argmin k^x q r s2s mbm sk 2. Nearest MBM signal mapping end for break; else j = j + 1 end if until j < K (N m 1) Output: The estimated MU-MBM signal vector ^x = [^x 1T ; ^x 2T ; ; ^x K T ] T

36 MU-MBM performance Bit error rate K = 2, n r = 8, 4 bpcu per user ML detection MU-CM, n t = 1,n rf = 1, 16-QAM MU-SM, n t = 2,n rf = 1, 8-QAM MU-GSM, n t = 4,n rf = 2, BPSK MU-MBM, n t = 1,n rf = 1,m rf = 3, BPSK SNR in db Figure : BER performance of MU-MBM, MU-CM, MU-SM, and MU-GSM with K = 2, n r = 8, 4 bpcu per user, and ML detection

37 MU-MBM performance Bit error rate MMSE Prop. Det. (OMP) Prop. Det. (CoSaMP) Prop. Det. (SP) K = 16, n t = 1, n r = 128 m rf = 6, 4-QAM 8 bpcu per user SNR in db Figure : BER performance of MU- a system with K = 16, n r = 128, n t = 1, n rf = 1, m rf = 6, 4-QAM, 8bpcu per user, using the proposed detection algorithm

38 MU-MBM performance Bit error rate MU-CM, nt = 1, nrf = 1, 32-QAM, MLD MM-GSM, nt = 5,nrf = 2, BPSK, Prop. Det. (SP) MU-SM, nt = 4,nrf = 1, 8-QAM, Prop. Det. (SP) MU-MBM, nt = 1,nrf = 1,mrf = 3, 4-QAM, Prop. Det. (SP) K = 16, n r = bpcu per user SNR in db Figure : BER performance MU-MBM, MU-CM, MU-SM, and MU-GSM in a setting with K = 16, n r = 128, and 5 bpcu per user

39 MU-MBM performance Bit error rate MU-CM, n t = 1, n rf = 1, 32-QAM, MLD MU-GSM, n t = 5,n rf = 2, BPSK, Prop. Det. (SP) MU-SM, n t = 4,n rf = 1, 8-QAM, Prop. Det. (SP) MU-MBM, n t = 1,n rf = 1,m rf = 3, 4-QAM, Prop. Det. (SP) 1-4 K = 16, 5 bpcu per user 1-5 SNR = 4 db n r Figure : BER performance MU-MBM, MU-CM, MU-SM, and MU-GSM as a function of n r in a setting with K = 16, 5 bpcu per user, and SNR = 4 db

40 uses digitally controlled parasitic elements (RF mirrors) for purposes conveys information by indexing RF mirrors oers rate, performance, hardware, and cost advantages suited for point-to-point and multiuser communications a promising approach for next generation wireless (5G)

41 Thank you