Media-based Modulation: Improving Spectral Efficiency beyond Conventional MIMO

Transcription

1 Media-based Modulation: Improving Spectral Efficiency beyond Conventional MIMO E. Seifi, M. Atamanesh A. K. Khandani E&CE Department, University of Waterloo 1

2 Basic Idea: Think of Smoked Signalling Channel State Antenna - Message is formed external to the TX antenna. - Message is encoded into selectable (pseudorandom) channel states.

3 A Major Difference with ConvenAonal Smoked Signalling Correspondence between messages and channel states is NOT pre- determined. TransmiNer is capable of selecong one out of M= R channel states. During a training phase channel states 1 to M are selected one aper the other, signifying messages 1 to M (as M constellaoon points) to the receiver. 3

4 Media- based Wireless Data Exponential growth due to rich scattering Carrier Keep the source shining and change the transmission medium to embed data. 4

5 Example: Using TradiAonal Ways Have two beams of gain 0.5 and 1.5 and have to use only one of them -Do not know which beam has higher gain. Want to form a 4 points 1-D constellation Option 1: Select one beam at random and use a 4 PAM constellation 50% chance of having a constellation with d min = 1/ d min = 1/ d min =1 Energy spent: 5/4 d min = 1/ 5

6 Example: A BeQer AlternaAve Have two beams of gain 0.5 and 1.5 and have to use only one of them -Do not know which beam has higher gain. Want to form a 4 points 1-D constellation Option : Select one bits of information to select one of the beams, and modulate a BPSK d min = 0 1 Energy spent:1/4 d min = d min =1 d min = 1 10 Times Energy Saving & Built-in Ergodicity 6

7 How to Change the Channel State? 14 RF Mirrors è 14 channel states è Modulate 14 bits 7

8 Examples of Antenna PaQerns 8

9 PropagaAon Environments Indoor Model (residential with dry-walls) Outdoor Model (down-town Ottawa) 9

10 Examples of ResulAng ConstellaAons 1 0

11 An Important Difference with MIMO Unlike MIMO, there is no fundamental restricaon on the physical size of the structure to embed the informaaon in the channel state. On the contrary, it is easier to affect the RF signal in near field. Why then using such a large structure? Did not know, and did not try much 11

12 Media- based vs. Legacy Transmission Main idea: Embed the informaaon in the variaaons of the channel state, external to the TX antenna. 1

13 Benefits vs. Legacy Transmission AddiAvity of informaaon over K receive antennas (similar to KxK MIMO) with the advantages: Unlike MIMO, K receive dimensions are realized using a single TX antenna Unlike MIMO, noise terms over the K orthogonal receive dimensions are independent of each other Inherent diversity over a staac channel using a single TX/RX antenna or mulaple TX/RX antennas Unlike MIMO, diversity does not require sacrificing the rate Transforms the staac Raleigh fading channel into AWGN channel 13

14 Basic Model m: Data Carrier m h(m) m =1,, L E h k (m) =1 z E( z k ) = σ y = h(m)+ z Vectors of Dimension K, K=Q is the # of Receive Antennas I( y;m) = I( y; h(m)) = H( y) H( z) = H( y) K log (πeσ ) h(m), m =1,, L : K-D constellation (iid Gaussian elements) 14

15 L as L result a As K h N q, 0 1, / σ Gain due to Inherent Diversity: Typicality of Random ConstellaAon 15 z z y + z k =1 N q m h(m) m =1,, L m: Data AWGN: Carrier of Energy E R R =. ) ( ] ) ( log 1 ) ( ) log( 1 [ )}] ( log 1 { ) log( 1 [ ) ( ) ( ), ( 1) ( x h x P L G h N h N N c G Y h N N C Y q d x e h x x f K L d h e c f e K e E e K z h N H y H I q q Q q σ σ σ π σ π σ σ π σ π Gaussian N m h y q ~ ) = ( + y

16 Main Conclusion Consider a staac Raleigh fading channel for which staasacal average of fading per receive antenna is one. Using a single TX antenna with one unit of energy, and Q RX antennas, mutual informaaon for a constellaaon with L points approaches the capacity of Q parallel SISO AWGN channels, each with unit energy, as L. It converges fast: σ 1 N q h 0, L It offers built- in error correcaon over a single transmission (no need for a powerful FEC). 16 / K as L

17 Media- based vs. MIMO KxK MIMO K complex Dimensions E/K E/K E/K H Total signal energy: KE Basis: Non-orthogonal Complex Dimensions/sec/Hz: K Better Performance Data Media-based one complex dimension E Total signal energy: KE Basis: Orthogonal Complex Dimension/sec/Hz: K 17

18 Main Benefit: Inherent Diversity in A Single ConstellaAon ConvenAonal methods suffer from deep fades in slow fading. This problem disappears as Good and Bad channel realizaaons contribute to forming the constellaaon. 18

19 Media- based vs. Legacy Systems: EffecAve Dimensionality λ 1 > λ >... > λ K : Eigenvalues of a KxK Wishart random matrix 1xK Media-based KxK legacy MIMO slope = λ 1 + λ + λ λ k E(λ 1 + λ + λ λ k ) = K E ( slope) = K slope = σ 1 E(λ 1 ) < λ λ 1 slope = λ 1 + λ slope = λ 1 + λ + λ 3 λ 3 ( 1 λ λ ) slope =1 K 1 λ k K 1 Legacy SISO Rate=log(1+E) 1 k=1 λ k E 19

20 PracAcal ConsideraAons Embedding large amount of data in the channel state, say 3 bits per channel use, is challenging: Training is difficult DetecAon is difficult Tracking is difficult SoluAon: Using a layered structure with mulaple transmit antennas. 0

21 Layered MIMO- MBM 1

22 Further SimplificaAon of MIMO- LMBM

23 Layered MIMO- MBM Training: N separate trainings, each with alphabet size R n, R n = R m Tracking: N separate trainings, each with alphabet size R n, R = R m n N N, N is the number of layers, N is the number of layers Detec-on: Can use an iteraave greedy scheme based on successive detecaon of layers. 3

24 Layered MIMO- MBM Relying on a layered structure contradicts the requirement of independence of code components in random coding. Important ObservaAon: DegradaAon in BER performance due to violaang the independence requirement in random coding is small. 4

25 Example of DegradaAon due to ViolaAng the Independence CondiAon 16 Bits Per Channel Use No Forward Error Correction 5

26 Example of BER Performance 3 Bits Per Channel Use No Forward Error Correction 6

27 Some Remarks Very low error in a single transmission Lowest possible delay Easy to track changes in the constellaaon structure Easy to perform equalizaaon using decision feedback 7

28 Coding Across ConsecuAve Time Transmissions GeneralizaAon of known binary codes Parity equaaons are generalized to modulo M, namely the constellaaon size A binary code with minimum distance d results in a minimum Ame diversity of order d. 8

29 Symbol- based Channel Coding Theorem: Under a set of mild condiaons, by applying FEC with error correcaon capability t, the slope of the error rate vs. SNR (with hard decision decoding) will asymptoacally increase by a factor of t + 1. Mimics the effect of diversity over Ame. 9

30 Example of Performance Gain due to Using Reed- Solomon Codes 30

31 SISO 3/5/7 Bits/S/HZ 31

32 1x8 16 bits/s/hz ~50dB 3

33 1x8 MBM 4x4 MIMO 33

34 1x8 MBM 8x8 MIMO 34

35 1x16 3 bits/s/hz 35

36 1x16 MBM 8x8 MIMO 36

37 1x16 MBM 4x16 MIMO 37

38 1x16 MBM 16x16 MIMO 38

39 Comparison with SpaAal ModulaAon 39

40 Thank You and QuesAons 40