31
On the SNR Exponent of Hybrid Digital Analog Space Time Coding Krishna R. Narayanan Texas A&M University http://ee.tamu.edu/~krn Joint work with Prof. Giuseppe Caire University of Southern California

On the SNR Exponent of Hybrid Digital Analog Space Time Coding

  • Upload
    lilith

  • View
    20

  • Download
    0

Embed Size (px)

DESCRIPTION

On the SNR Exponent of Hybrid Digital Analog Space Time Coding. Krishna R. Narayanan Texas A&M University http://ee.tamu.edu/~krn Joint work with Prof. Giuseppe Caire University of Southern California. s. s. s. K. 1. 2. ;. ;. :. :. :. ;. ^. ^. s. s. K. 1. ;. :. :. :. ;. - PowerPoint PPT Presentation

Citation preview

Page 1: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

On the SNR Exponent ofHybrid Digital Analog Space Time Coding

Krishna R. NarayananTexas A&M University

http://ee.tamu.edu/~krn

Joint work with Prof. Giuseppe CaireUniversity of Southern California

Page 2: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

Page 3: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

What this talk is about

Transmit an analog source over a MIMO channel

Encoder

s1;s2; : : : ;sK s1; : : : ; sKReceiver

Quasi-static (Block) Fading

Performance criterion : End to End quadratic distortion

How to best use multiple antennas to minimize distortion?

T uses of the channel

Page 4: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

Example 1: Streaming real-time video

Due to stringent delay constraints, we cannot code over many realizations of the channel

Channel changes from one block to another

Encoder

s1;s2; : : : ;sK s1; : : : ; sKReceiver

Quasi-static (Block) Fading

Page 5: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

Example 2: Broadcasting an analog source

Empirical distribution of Hl converges to some ensemble distribution as l ! 1

Encoder

s1;s2; : : : ;sK

s1; : : : ; sKUser 1

s1; : : : ; sKUser 2

s1; : : : ; sKUser 3

H 1

H 2

H 3

Page 6: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

Block Fading Channel Model

yt =p

½M H xt + wt; t = 1;: : : ;T

½denotes theSignal-to-Noise Ratio (SNR)

X = [x1; : : : ;xT ], is normalized such that tr(E[X HX ]) · M T

T is theduration (in channel uses)

M · N (the case M > N follows as a simple extension)

H 2 CN £ M , wherehi ;j » CN (0;1)

Page 7: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

Diversity Multiplexing Tradeoff

Channel has zero capacity, we talk of outage capacity

For Pe() ! 0, we need SNR ! 1

Makes sense to look at the exponent of SNR

Diversity gain : Pe() / -d

Multiplexing gain: Use many antennas to transmit many data streams

Zheng and Tse: Both are simultaneously achievable, there is a fundamental trade off of how much of each we can achieve

Page 8: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

Optimal D-M Tradeoff (Zheng and Tse)

Consider a family of space-time coding schemes fCr c(½)g

Rc = rc log½ rc is the muliplexing gain

Optimal exponent d¤(rc) = supd(rc)

whereas rate increases with ½as r = rc log½

For large ½, Poutage / ½¡ d(r c )

Page 9: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

d*(r) for Rayleigh Fading Channels

d¤(r) = (M ¡ r) (N ¡ r)

Page 10: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

Precise Problem Statement

SourceChannelEncoder

s1;s2; : : : ;sK s1; : : : ; sKReceiver

Fadingsi » N (0;1)

D(½) ¢= 1

KE[js ¡ esj2]

Family of source-channel coding schemes fSC (½)g

Rate of decay of MSE with SNR: D / ½¡ a(´)

SNR Exponent: a(´) = ¡ lim½! 1logD (½)

log½

a?(´) = supa(´)

Channel

SNR = ½

´ = 2KT

Page 11: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

Exponent for Separation Scheme

Sp-TimeCode

Q MIMOChannel

Rs = rs log½bits Rc = rc log½bits

SinceRsK = RcT, wehaveRc = ´Rs

Equating exponents, we get the result in Theorem 1

Dsep(Rs) · DQ (Rs) +a P (e)

Dsep(Rs) · ½¡ 2´ rc +a ½¡ d¤ (r c )

Not in outage In outage

Page 12: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

Page 13: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

Main Results – Separation based approach

Comments – Separation is not optimal

Can we outperform the optimized separated scheme ? Yes, we will see that later

T heorem 1 [Exponent achievable by separation]

asep(´) = 2(j d? (j ¡ 1)¡ (j ¡ 1)d? (j ))2+´(d? (j ¡ 1)¡ d? (j )) ´ 2

h2(j ¡ 1)d? (j ¡ 1) ;

2jd? (j )

´for j = 1;:: : ;M ,

is achievable by a tandemsource-channel coding scheme ¤

Page 14: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

Upper bound on the exponent

An upper bound is obtained by assuming that the channel is instantaneously known at the transmitter

Coding rate is chosen according to

Large SNR behavior can be analyzed using techniques similar to those in Zheng and Tse (Wishart distribution, Varadhan’s lemma)

Rc(H ) · logdet¡I + ½HH H

¢

D(½) ¸ Eh

1det(I +½H H H)2=´

i

Page 15: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

Upper bound

T heorem 2 [Informed transmitter upper bound]

The optimal distortion SNR exponent a?(´) is upperbounded by

aub(´) =MX

i=1

min½

;2i ¡ 1+ jM ¡ N j¾

(1

¤

Page 16: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

The case for analog coding schemes – known SNR

n1; : : : ;nk

Channelcode

Q + Decoder

s1; : : : ;sk

OR

s1; : : : ;sk +MMSE

estimate

n1; : : : ;nk

rs = 12 log(1+ ½) D(½) = 1

1+½

D(½) = 11+½

Page 17: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

Separation scheme with Fading

n1; : : : ;nk

Channelcode

Q + Decoder

s1; : : : ;sk

Separation based scheme is optimal only when ||hi||2 = 1

Can be quite bad otherwise

Cannot exploit higher instaneous channel gain

rs = 12 log(1+ ½)

hi

Page 18: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

Analog scheme with Fading

s1; : : : ;sk +MMSE

estimate

n1; : : : ;nk

Analog scheme is simultaneously optimal for all SNRs

Graceful degradation of MSE with SNR

hi

D(½;hi ) = 11+jjhi jj2½

Page 19: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

Why Hybrid then?

Alas! things are never that easy

The optimality is valid only for the SISO channel with T = K

If more bandwidth is available, it is difficult to take advantage

Same with lesser bandwidth also

Hence, we need to look for hybrid digital and analog solutions

Page 20: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

HDA Solution for T > K (Bandwidth expansion)

Involves some math, but the exponent can be analyzed

Express MSE as a function of the Eigen values, use the Wishart distribution and Varadhan’s lemma

Quantizer

K rs log½bits

- Reconstruct

Space-TimeEncoder

X (d) 2 CM £ Td

SpatialMultiplexer

X (a) 2 CM £ K2M

s1; : : : ;sK

s1; : : : ; sK

e1; : : : ;eK

Page 21: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

QAM

Page 22: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

Exponent of Hybrid Digital Analog Coding Schemes

This is better than the separated exponent

T heorem 3 [H ybrid scheme lower bound]

For BW expansion, thehybrid digital-analog (HDA) space-timecoding achieves

ahybrid(´) = 1+µ

¡1M

¶j d?(j ¡ 1) ¡ (j ¡ 1)d?(j ) ¡ 1

2´ ¡ 1

M + d?(j ¡ 1) ¡ d?(j )(1)

Page 23: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

HDA Scheme for T < K (Bandwidth Compression)

QuantizerSpace-Time

Encoder

SpatialMultiplexer

+

X (D ) 2 CM £ T

X (a) 2 CM £ T

1 11,..., Ks s

1,...,K Ks s

1 logsK r bits

Main trick is in chosing ¯ to be dependent on SNR as ½¡ °

and optimizing °

Page 24: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

Page 25: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

Exponent of HDA Schemes

It is remarkable that this is equal to the upper bound

Hence it is optimal

T heorem 3 [Hybrid scheme lower bound]

Bandwidth Compression:

ahybrid(´) =2M´

; ´ ¸ 2M

Page 26: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

Upper bound for the Parallel Channel

For M parallel channels

aub = min(M; M2´ )

Page 27: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

Scalar channel M = N = 1

Page 28: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

MIMO M = N = 2

Page 29: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

Comments

SISO Channel with fading

Even if T > K, the best exponent is 1

More bandwidth does not buy us anything

It is the degrees of freedom, not the bandwidth that is important

For this case, the exponent for the entire region is fully known

Gunduz-Erkip - infinite layers of superposition coding is optimal

Our approach is much simpler

Page 30: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

MIMO Case

However, in the MIMO case things are different

More bandwidth helps us buy diversity

Antennas can be used to compress

It is very easy to find practical schemes to get the correct exponent

For example, uniform scalar quantization is optimal at high SNRs!

We may require very large SNR for the asymptotics to kick in

Page 31: On the SNR Exponent of Hybrid Digital Analog Space Time Coding

Wireless Communications Lab, TAMU

Outlook

Our conjecture is that the upper bound is loose (it is not entirely clear) for the bandwidth expansion case

Better constructive schemes to improve the achievable part

In some sense, the key problem is to find a better exponent for the parallel channel