discrete signaling for non-coherent, single-antenna

1
Discrete Signaling for Non-Coherent, Single-Antenna, Rayleigh Block-Fading Channels Marcin Pikus 12 , Gerhard Krammer 2 , and Georg B¨ ocherer 2 1 Huawei Technologies Duesseldorf GmbH, Germany 2 Institute for Communications Engineering, Technical University of Munich, Germany Rayleigh Block Fading Channel (RBFC) Used in modeling of wireless channels — it captures the uncertainty of wireless channels and fadings correlation-in-time The fading coefficient H is constant within a block of T symbols and changes independently between blocks The parameter T represents the channel coherence time The fading coefficient H is unknown to the transmitter and receiver The receiver has to estimate the channel state and the data from the received symbols γ X × + Y H ∼N C (0, 1) N ∼N C (0 , I) E X 2 = T [Y 1 ,..., Y T ] = H · γ ·[X 1 ,..., X T ] +[N 1 ,..., N T ] (1) Y = H · γ X + N (2) Signaling in the Literature CSI Capacity When the receiver knows H , Gaussian IID input signal X ∼N C (0 , I) achieves the capacity. This rate upper-bounds all non-coherent rates. Gaussian IID When the receiver does not know H , Gaussian IID input signal X ∼N C (0 , I) in not capacity-achieving. However it performs well for large T [1]. Pilot schemes The transmitter inserts a fixed pilot symbol inside each fading block. At the receiver the pilot symbol is used to estimate the fading H and then the estimate ˆ H is used to decode the data [2]. USTM Unitary Space-Time Modulation transmits in each fading block a vector Φ uniformly distributed on the T -dimensional complex sphere with radius T [3]. Experimental Signaling Discrete Product Form is the proposed discrete signaling scheme inspired by the capacity achieving distribution for RBFC [4]. On-Off USTM is an extension of the USTM. The scheme transmits either a vector Φ as in USTM or a vector of 0’s. We optimize the probability P 0 of transmitting the 0 -symbol. This scheme is an isotropicaly (continuous) counterpart of DPF. On-Off Gaussian is an extension of Gaussian IID signaling. The scheme transmits either a vector of Gaussian IID RVs or a vector of 0’s. We optimize the probability P 0 of transmitting the 0 -symbol. The Capacity-Achieving Coding Scheme The capacity-achieving signaling over non-coherent RBFC (also called product form) reads as X = V Φ (3) V and Φ are independent RVs Φ is uniformly distributed on the complex T -dimensional sphere with radius T V is discrete real non-negative RV v 1 Φ 1 v 1 Φ 2 v 2 Φ 3 v 2 Φ 4 Discretization 1 Choose spheres radius, i.e., choose RV V V = 0, with probability 1 - P 1 v 1 = 1 P 1 , with probability P 1 (4) 2 Sample the spheres by choosing the following points Θ = [Θ 1 ,..., Θ T ] (5) where Θ i are IID uniformly distributed RVs on the QPSK alphabet A 4 = {1, -1, j , -j }. We get the Discrete Product Form scheme X = 0 , with probability 1 - P 1 1 P 1 Θ with probability P 1 , (6) and P 1 is chosen to maximize the mutual information. Computing the Mutual Information The following expansion will be used to compute the mutual information I (Y ; X )= h(Y ) - h(Y |X ) (7) The RV Y |{X = x } is a zero-mean circular-symmetric Gaussian RV. The covariance matrix equals to E Y Y X = x = I n B + γ x x (8) Using the log-det formula, the entropy h(Y |X ) is h(Y |X )= E X [h(Y |X = X )] = T log(πe )+ P 1 log 1+ γ T P 1 . Results: Rate vs Eb/N0 Computing h(Y ) The following formula can be used to compute h(Y ) h(Y )= E Y [- log p(Y )]. (9) The knowledge of p(y ) is needed to evaluate (9). However, a direct computation fails due to exponential complexity growth in block length T p(y )= x P (x )p(y |x ) (10) Instead, we use the fact that Y conditioned on V and H is a vector of IID RVs. First expand p(y ) = Pr(V =0) p(y |V =0) ∼N C (0 , I) + Pr(V = v 1 )p(y |V = v 1 ) (11) Then the second term p(y |v 1 )= E H p(y |v 1 , H ) = E H T i =1 p(y i |v 1 , H ) = E H T i =1 θ i ∈A 4 P (θ i )p(y i |θ i , v 1 , H ) = E H T i =1 θ i ∈A 4 1 4 · 1 π e -|y i -H γθ i v 1 | 2 = E H f y , v 1 , H (12) The complexity of computing f y , v 1 , H grows linearly with respect to the block size T . To evaluate the expectation in (12) we apply MC averaging. Discussion/Conclusions Gap to continuous dist.: The proposed Discrete Product Form (light green) achieves similar performance to its isotropical counterpart (On-Off USTM, blue) up to 0.4 b/CU and saturates approx at 1 b/CU and 2 b/CU for block lengths T = 2 and T = 50, respectively. Using the 0 -symbol: Compare the performance of the corresponding schemes: USTM (dark green) and On-Off USTM (blue) or Gaussian IID (red) and On-Off Gaussian (brown). Introducing a non-zero probability mass at the 0 -symbol significantly improves the performance. Also for higher SNR, especially with small T . Using on-a-sphere signal: For T = 2 observe the gap between On-Off Gaussian (brown) and On-Off USTM (blue). Both signals are on-off schemes so the gain is due to the on-a-sphere structure of the On-Off USTM during th on-cycle. For low rates, also Discrete Product Form (light green) is better than On-Off Gaussian for the same reason. References [1] F. Rusek, A. Lozano, and N. Jindal, “Mutual information of IID complex Gaussian signals on block Rayleigh-faded channels,” IEEE Trans. Inf. Theory, vol. 58, no. 1, pp. 331–340, Jan. 2012. [2] B. Hassibi and B. Hochwald, “How much training is needed in multiple-antenna wireless links?” IEEE Trans. Inf. Theory, vol. 49, no. 4, pp. 951–963, Apr. 2003. [3] B. Hochwald and T. Marzetta, “Unitary space-time modulation for multiple-antenna communications in Rayleigh flat fading,” IEEE Trans. Inf. Theory, vol. 46, no. 2, pp. 543–564, Mar. 2000. [4] M. Pikus, G. Kramer, and G. Bocherer, “Discrete signaling for non-coherent, single-antenna, rayleigh block-fading channels,” IEEE Communications Letters, vol. PP, no. 99, pp. 1–1, 2016. Institute for Communications Engineering Technische Universit¨ at M¨ unchen 0 2 4 6 8 10 0 0.2 0.4 0.6 0.8 1 1.2 1.4 E b /N0 [dB] Rate [b/CU] CSI Capacity On-Off USTM Disc. Prod. Form Gaussian IID On-Off Gaussian USTM Pilot Scheme -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 3 3.5 0 0.2 0.4 0.6 0.8 1 1.2 1.4 On-Off USTM Gaussian IID On-Off Gaussian USTM Gaussian IID USTM E b /N0 [dB] Rate [b/CU] CSI Capacity On-Off USTM Disc. Prod. Form Gaussian IID On-Off Gaussian USTM Pilot Scheme Information rates for block length T = 2. Information rates for block length T = 50.

Upload: others

Post on 25-Jan-2022

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Discrete Signaling for Non-Coherent, Single-Antenna

Discrete Signaling for Non-Coherent, Single-Antenna, Rayleigh Block-Fading ChannelsMarcin Pikus12, Gerhard Krammer2, and Georg Bocherer2

1Huawei Technologies Duesseldorf GmbH, Germany 2Institute for Communications Engineering, Technical University of Munich, Germany

Rayleigh Block Fading Channel (RBFC)

Used in modeling of wireless channels — it captures theuncertainty of wireless channels and fadings correlation-in-time

The fading coefficient H is constant within a block of Tsymbols and changes independently between blocksThe parameter T represents the channel coherence timeThe fading coefficient H is unknown to the transmitter andreceiverThe receiver has to estimate the channel state and the datafrom the received symbols

√γX × + Y

H ∼ NC (0, 1) N ∼ NC (0, I)

E[‖X‖2] = T

[Y1, . . . ,YT ]† = H·√γ·[X1, . . . ,XT ]† + [N1, . . . ,NT ]† (1)

Y = H·√γX + N (2)

Signaling in the Literature

CSI Capacity When the receiver knows H, Gaussian IID inputsignal X ∼ NC (0, I) achieves the capacity. This rateupper-bounds all non-coherent rates.

Gaussian IID When the receiver does not know H, Gaussian IIDinput signal X ∼ NC (0, I) in not capacity-achieving.However it performs well for large T [1].

Pilot schemes The transmitter inserts a fixed pilot symbol insideeach fading block. At the receiver the pilot symbol isused to estimate the fading H and then the estimate H isused to decode the data [2].

USTM Unitary Space-Time Modulation transmits in each fadingblock a vector Φ uniformly distributed on theT -dimensional complex sphere with radius

√T [3].

Experimental Signaling

Discrete Product Form is the proposed discrete signaling schemeinspired by the capacity achieving distribution for RBFC[4].

On-Off USTM is an extension of the USTM. The schemetransmits either a vector Φ as in USTM or a vector of0’s. We optimize the probability P0 of transmitting the0-symbol. This scheme is an isotropicaly (continuous)counterpart of DPF.

On-Off Gaussian is an extension of Gaussian IID signaling. Thescheme transmits either a vector of Gaussian IID RVs ora vector of 0’s. We optimize the probability P0 oftransmitting the 0-symbol.

The Capacity-Achieving Coding SchemeThe capacity-achieving signaling over non-coherent RBFC (alsocalled product form) reads as

X = V Φ (3)

V and Φ are independent RVsΦ is uniformly distributed on the complex T -dimensionalsphere with radius

√T

V is discrete real non-negative RV

v1Φ1

v1 Φ2

v 2Φ3

v2Φ4

Discretization

1 Choose spheres radius, i.e., choose RV V

V =

0, with probability 1− P1

v1 =√

1P1, with probability P1

(4)

2 Sample the spheres by choosing the following points

Θ = [Θ1, . . . ,ΘT ] (5)

where Θi are IID uniformly distributed RVs on the QPSKalphabet A4 = {1,−1, j ,−j}.

We get the Discrete Product Form scheme

X =

0, with probability 1− P1√1

P1Θ with probability P1,

(6)

and P1 is chosen to maximize the mutual information.

Computing the Mutual Information

The following expansion will be used to compute the mutualinformation

I(Y ; X ) = h(Y )− h(Y |X ) (7)

The RV Y |{X = x} is a zero-mean circular-symmetricGaussian RV. The covariance matrix equals to

E[Y Y †

∣∣∣X = x]

= InB + γxx † (8)

Using the log-det formula, the entropy h(Y |X ) is

h(Y |X ) = EX [h(Y |X = X )]

= T log(πe) + P1 log(

1 + γTP1

).

Results: Rate vs Eb/N0

Computing h(Y )

The following formula can be used to compute h(Y )

h(Y ) = EY [− log p(Y )]. (9)

The knowledge of p(y) is needed to evaluate (9). However, adirect computation fails due to exponential complexity growthin block length T

p(y) =∑

xP(x)p(y |x) (10)

Instead, we use the fact that Y conditioned on V and H is avector of IID RVs. First expand

p(y) = Pr(V=0) p(y |V=0)︸ ︷︷ ︸∼ NC (0, I)

+ Pr(V=v1)p(y |V=v1) (11)

Then the second term

p(y |v1) = EH[p(y |v1,H)

]= EH

[ T∏i=1

p(yi |v1,H)]

= EH

T∏i=1

∑θi∈A4

P(θi )p(yi |θi , v1,H)

= EH

T∏i=1

∑θi∈A4

14 ·

e−|yi−H√γθi v1|2

= EH[f(y , v1,H

)](12)

The complexity of computing f(y , v1,H

)grows linearly with

respect to the block size T . To evaluate the expectationin (12) we apply MC averaging.

Discussion/Conclusions

Gap to continuous dist.: The proposed Discrete Product Form(light green) achieves similar performance to itsisotropical counterpart (On-Off USTM, blue) up to 0.4b/CU and saturates approx at 1 b/CU and 2 b/CU forblock lengths T = 2 and T = 50, respectively.

Using the 0-symbol: Compare the performance of thecorresponding schemes: USTM (dark green) and On-OffUSTM (blue) or Gaussian IID (red) and On-Off Gaussian(brown). Introducing a non-zero probability mass at the0-symbol significantly improves the performance. Also forhigher SNR, especially with small T .

Using on-a-sphere signal: For T = 2 observe the gap betweenOn-Off Gaussian (brown) and On-Off USTM (blue).Both signals are on-off schemes so the gain is due to theon-a-sphere structure of the On-Off USTM during thon-cycle. For low rates, also Discrete Product Form(light green) is better than On-Off Gaussian for the samereason.

References

[1] F. Rusek, A. Lozano, and N. Jindal, “Mutual information of IID complexGaussian signals on block Rayleigh-faded channels,” IEEE Trans. Inf. Theory,vol. 58, no. 1, pp. 331–340, Jan. 2012.

[2] B. Hassibi and B. Hochwald, “How much training is needed inmultiple-antenna wireless links?” IEEE Trans. Inf. Theory, vol. 49, no. 4, pp.951–963, Apr. 2003.

[3] B. Hochwald and T. Marzetta, “Unitary space-time modulation formultiple-antenna communications in Rayleigh flat fading,” IEEE Trans. Inf.Theory, vol. 46, no. 2, pp. 543–564, Mar. 2000.

[4] M. Pikus, G. Kramer, and G. Bocherer, “Discrete signaling for non-coherent,single-antenna, rayleigh block-fading channels,” IEEE CommunicationsLetters, vol. PP, no. 99, pp. 1–1, 2016.

Institute forCommunications Engineering Technische Universitat Munchen

0 2 4 6 8 100

0.2

0.4

0.6

0.8

1

1.2

1.4

Eb/N0 [dB]

Rate

[b/C

U]

CSI CapacityOn-Off USTMDisc. Prod. FormGaussian IIDOn-Off GaussianUSTMPilot Scheme

−1.5 −1 −0.5 0 0.5 1 1.5 2 2.5 3 3.50

0.2

0.4

0.6

0.8

1

1.2

1.4

On-Off USTMGaussian IIDOn-Off GaussianUSTM

Gaussian IIDUSTM

Eb/N0 [dB]

Rate

[b/C

U]

CSI CapacityOn-Off USTMDisc. Prod. FormGaussian IIDOn-Off GaussianUSTMPilot Scheme

Information rates for block length T = 2. Information rates for block length T = 50.