SPECTRE

Short Packet Communication Toolbox

Version 0.2

Contributors (alphabetic order):

Austin Collins, Giuseppe Durisi, Tomaso Erseghe, Victoria Kostina,
Johan Ostman, Yury Polyanskiy, Ido Tal, Wei Yang

e-mail: [email protected]

September 2, 2016

Abstract

Shannon theory describes fundamental limits of communication and

compression systems. Classic closed-form results (such as the well known
log(1 + SNR) formula) apply only to the regime of infinite blocklength
(infinite packet size/delay). For finite blocklengths, no closed-form results
are usually obtainable, but there exist tight upper and lower bounds, as
well as approximations. This manual describes numerical routines for
computing these bounds and approximations for some popular channel
and source models.
The toolbox is under development and the participation of additional
members of the information and communication theory communities in
this endeavor is warmly welcomed!

1
Contents

Contents 2

1 Motivation 5

2 The real AWGN channel 7

2.1 Summary plot: plot v3() . . . . . . . . . . . . . . . . . . . . . 7
2.2 Achievability: shannon ach2() . . . . . . . . . . . . . . . . . . 8
2.3 Achievability: kappabeta ach() . . . . . . . . . . . . . . . . . 8
2.4 Achievability: gallager ach() . . . . . . . . . . . . . . . . . . 8
2.5 Converse: converse() . . . . . . . . . . . . . . . . . . . . . . . 9
2.6 Normal approximation: normapx awgn(), normapx biawgn() . 9
2.7 Code database: plot universe() . . . . . . . . . . . . . . . . 9

3 Block-memoryless Rayleigh fading (no CSI) 11

3.1 Achievability bounds . . . . . . . . . . . . . . . . . . . . . . . . 12
3.2 Converse helper functions . . . . . . . . . . . . . . . . . . . . . 12
3.3 Converse bound . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.4 Alamouti . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.5 Outage capacity . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.6 Ergodic capacity . . . . . . . . . . . . . . . . . . . . . . . . . . 14

4 Quasi-static fading channel 15

4.1 Summary plot: plot all() . . . . . . . . . . . . . . . . . . . . 16
4.2 Achievability: ach simo csir() . . . . . . . . . . . . . . . . . . 16
4.3 Achievability: ach simo nocsi() . . . . . . . . . . . . . . . . . 17
4.4 Achievability: mimo iso ach() . . . . . . . . . . . . . . . . . . 17
4.5 Converse: converse simo() . . . . . . . . . . . . . . . . . . . . 17
4.6 Converse: mimo iso conv() . . . . . . . . . . . . . . . . . . . . 17
4.7 Normal approximation . . . . . . . . . . . . . . . . . . . . . . . 17

5 Quasi-static fading channel (long-term power constraint) 19

5.1 The AWGN channel (H = 1) . . . . . . . . . . . . . . . . . . . 20
5.2 Quasi-static fading channel with perfect CSIT . . . . . . . . . . 20

6 Block-memoryless Rayleigh Fading (CSIR) 21

6.1 Summary plot: plot all() . . . . . . . . . . . . . . . . . . . . 22
6.2 Achievability Bound: MIMO achievability CSIR() . . . . . . . 22
6.3 Normal Approximation: MIMO csir normapx() . . . . . . . . . 23

2
CONTENTS 3

7 Energy-per-bit under vanishing spectral efficiency 25

7.1 Summary plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
7.2 The AWGN channel (Hj = 1) . . . . . . . . . . . . . . . . . . . 26
7.3 Rayleigh fading: CSIR case . . . . . . . . . . . . . . . . . . . . 27
7.4 Rayleigh fading: noCSI case . . . . . . . . . . . . . . . . . . . . 27

8 Fixed-length source coding under a distortion criterion 29

8.1 Binary memoryless source . . . . . . . . . . . . . . . . . . . . . 30
8.2 Gaussian memoryless source . . . . . . . . . . . . . . . . . . . . 31
8.3 Binary erased source . . . . . . . . . . . . . . . . . . . . . . . . 32

9 Variable-length compression allowing errors 35

9.1 Binary memoryless source . . . . . . . . . . . . . . . . . . . . . 36

10 Joint source-channel coding under a distortion criterion 37

10.1 Binary memoryless source transmitted over a binary symmetric
channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
10.2 Gaussian source transmitted over an AWGN channel . . . . . . 39

11 The binary-symmetric channel (BSC) 41

11.1 Summary plot: plot all() . . . . . . . . . . . . . . . . . . . . 41
11.2 Achievability: dt ach() . . . . . . . . . . . . . . . . . . . . . . 42
11.3 Achievability: rcu ach() . . . . . . . . . . . . . . . . . . . . . 42
11.4 Achievability: gallager ach() . . . . . . . . . . . . . . . . . . 42
11.5 Converse: converse() . . . . . . . . . . . . . . . . . . . . . . . 42
11.6 Normal approximation: normapx() . . . . . . . . . . . . . . . . 43

12 The binary-erasure channel (BEC) 45

12.1 Summary plot: plot all() . . . . . . . . . . . . . . . . . . . . 45
12.2 Achievability: dt ach() . . . . . . . . . . . . . . . . . . . . . . 46
12.3 Achievability: gallager ach() . . . . . . . . . . . . . . . . . . 46
12.4 Converse: converse() . . . . . . . . . . . . . . . . . . . . . . . 46
12.5 Normal approximation: normapx() . . . . . . . . . . . . . . . . 47

13 The binary-AWGN channel (BI-AWGN) 49

13.1 Converse: converse mc() . . . . . . . . . . . . . . . . . . . . . 49
13.2 Example 1: example speff.m . . . . . . . . . . . . . . . . . . . 50
13.3 Example 2: example per.m . . . . . . . . . . . . . . . . . . . . 51

Bibliography 53
Chapter 1

Motivation

In his 1948 landmark paper, Claude E. Shannon demonstrated that communi-

cation with arbitrarily small probability of error is feasible if, and only if, the
communication rate is below the so called channel capacity. To achieve this
result, codes with large packet size (blocklength) must be employed. Shannons
channel capacity has served as a useful guideline to design wireless communica-
tion systems.
In some applications, however, a more refined analysis of the interplay be-
tween packet-error probability, communication rate, and packet size is required.
This may occur in emerging applications, such as massive machine-to-machine
communication for metering, traffic safety, and telecontrol of industrial plants,
together with real-time data transfer to enable remote wireless control (tactile
internet), which may require the exchange of short packets, sometimes under
stringent latency and reliability constraints.
Finite-blocklength information theory, a subfield of information theory that
benefitted from seminal contributions from Shannon, Strassen, and Dobrushin,
among others, is currently a very active field of research. Nonasymptotic
achievability and converse bounds on the maximum coding rate achievable for
a given packet size and packet error probability are now available for several
channel models that are relevant for wireless communication systems, such as
the AWGN channel, the quasi-static Rayleigh fading channel, and the Rayleigh
block-fading channel. The purpose of this toolbox is to provide numerical
routines for the computation of these bounds. In the next chapters, we provide
a bare-bone manual, which describes the routines available in this toolbox.

5
Chapter 2

The real AWGN channel

General info:
Maintainer: Y. Polyanskiy <[email protected]>
Main references: [1]
Example: See Fig. 2.1
Channel model:
Yj = Xj + Zj , j = 1, . . . , n
Codewords are real: xj R, j = 1, . . . , n
Noise: Zj N (0, 1) iid.

Power constraints: Each codeword xn satisfies

n
X
|xj |2 nP
j=1

Common input/output arguments:

1. P input argument; parameter of the power constraint (so P = 10
is SN R = 10 dB).
2. epsil block probability of block error.
3. lm or Lms output argument; log2 M , log-size of codebook (base-2
information units).
4. n input argument; blocklength.
For slow functions that do not support vectorized arguments.
5. Ns input argument; vector of blocklengths.

2.1 Summary plot: plot v3()

Definition:
function [Ns lb ub feinst gal wlfz] = plot_v3(P, epsil)
This function plots various bounds on the same figure. See Fig. 2.1.

7
8 CHAPTER 2. THE REAL AWGN CHANNEL

> plot_v3(10, 0.001, 1, 100:20:1000)

Bounds for AWGN, SNR = 10 dB, Pe = 0.001

1.8

1.6

1.4

1.2

Capacity
Rate, R

1
Converse (|X|2 = nP)
Normal approximation
0.8 kappabeta achievability
Gallager random coding

0.6

0.4

0.2

0
0 100 200 300 400 500 600 700 800 900 1000
Blocklen, n

Figure 2.1: Code and resulting picture for AWGN bounds

2.2 Achievability: shannon ach2()

Definition:

function lm = shannon_ach2(n, epsil, P)

Function computes Shannons cone-packing achievability bound, see [1, (41)].

2.3 Achievability: kappabeta ach()

Format:

function Lms = kappabeta_ach(Ns, epsil, P, hack)

This computes the achievability bound, see left-hand inequality in [1,

(218)]. If hack =
6 0, then we use asymptotic approximation for . This increases
the speed dramatically and is very precise already at n = 10, see [1, (217)]. If
hack is not set, it defaults to 1.

2.4 Achievability: gallager ach()

Format:

function lm = gallager_ach(n, epsil, P)

Computes Gallagers achievability bound, see [1, (44)].

2.5. CONVERSE: CONVERSE() 9

2.5 Converse: converse()

Format:

function Lms = converse(Ns, epsil, P)

This computes the meta-converse lower bound with QY n = N (0, 1 + P )n ,
see right-hand inequality in [1, (218)].

2.6 Normal approximation: normapx awgn(),

normapx biawgn()
Format:

function Lms = normapx_awgn(Ns, epsil, P);

Fast and frequently very precise approximation to both achievability and

converse:
1
log M nC nV Q1 () + log n
2

2.7 Code database: plot universe()

Code in universe/plot_universe() compiles a comparative plot of normalized
rate for different codes. See [1, Section IV.D].
Chapter 3

Block-memoryless Rayleigh
fading channel, no CSI

General info:

Maintainer: G. Durisi <[email protected]>

Contributors: G. Durisi, J. Ostman, W. Yang

Main reference: [2]

Example: See Fig. 3.1.

Channel model: Rayleigh block-fading channel with mt transmit antennas

and mr receive antennas that stays constant for nc channel uses. The
blocklength n is an integer multiple of the coherence time nc , i.e., n = lnc ;
here l denotes the number of independent time-frequency branches:

Yk = Xk Hk + Wk , k = 1, . . . , l , (3.1)
where Yk , Wk Cnc mr , Xk Cnc mt and Hk Cmt mr . The noise
Wk and the channel gain Hk are independent and have independent
CN (0, 1) entries. Neither the transmitter nor the receiver are aware of
the realizations of the fading matrix Hk .
The power constraint is defined per coherence interval, i.e., each codeword
C consists of l subcodewords C1 , . . . , Cl that must satisfy:

tr CH

k Ck = nc , k = 1, . . . , l.

Here, denotes the SNR.

Common input, output arguments:

snrdB SNR in dB
T channel coherence time (expressed in channel uses)
L number of independent fading realizations spanned by each
codeword
Mt number of tranmsmit antennas

11
12 CHAPTER 3. BLOCK-MEMORYLESS RAYLEIGH FADING (NO CSI)

coherence interval nc (log scale)

168 84 42 24 12 8 6 4
Cerg () (approx.)
MC upper bound
3 DT lower bound
R (nc , l, , )
2.5 Cout (, )
bit/channel use

1.5

1 Alamouti

0.5

0
1 2 4 7 14 21 28 42
time-frequency diversity branches l (log scale)

Figure 3.1: Maximum coding rate R (nc , l, , ) for 22 MIMO system operating
over a Rayleigh block-fading channel with coherence time nc . Each packet
spans l coherence intervals. The packet size is 168 channel uses. The packet
error rate is  = 103 . The SNR is = 6 dB.

Mr number of receve antennas

epsilon packet error rate (block error rate)
prec log2 of the number of samples used in the Monte-Carlo step;
this parameter should be set so that 2prec  100/epsilon
filename data file in which the samples of the information density
are saved for possible future refinements (provided that the flag SAVE
is set)
R estimate of the maximum coding rate
current_eps actual bound on the packet error rate (it may deviate
from epsilon if prec is not chosen appropriately)

3.1 Achievability bounds: DT USTM NxM(), DT USTM MxM()

Format:
function [R,current_eps]
=DT_USTM(snrdB,T,L,Mt,Mr,epsilon,prec,filename);
This function computes the USTM-DT achievability bound [2, Th. 3] for the
case mt mr .

3.2 Converse helper functions: MC USTM 2x2 q1x2(),

MC USTM 2x2 q 2x2(), MC USTM 4x4 q Mx4
Format
3.3. CONVERSE BOUND 13

function [R,current_eps,current_prec]
=MC_USTM_2x2(snrdB,T,L,Mtalt,epsilon,prec,pow_all,filename)
function [R,current_eps,current_prec]
=MC_USTM_4x4(snrdB,Mtalt,T,L,epsilon,prec,pow_all,filename)

These two functions compute the right-hand side of [2, Eq. (41)] for fixed input
diagonal matrices {k }lk=1 and fixed auxiliary output distribution (which is
parameterized by the number of effectively used transmit antennas Mtalt that
result in this distribution), for the case mt = mr = 2 and mt = mr = 4. To
obtain the actual converse bound, a further optimization over these parameters
need to be performed. This is discussed in the next section. The parameter
pow_all is related to the diagonal entries of the matrices {k }lk=1 . For the
mt = mr = 2 case, pow_all is an l-dimensional vector, whose entries are in
[0, 0.5]. Indeed, the particular formulation of the power constraint in [2, Eq. (41)]
implies that the other entry is uniquely determined from the first one. For the
4 4 case, the code provided assumes that the optimization in [2, Eq. (41)]
is restricted only to {k }lk=1 of the form 1 = = l = . In this case,
pow_all is a 4-dimensional vector that contains the diagonal entries of .

3.3 Converse bound: compute MC 2x2 telatar

Format

function R=compute_MC_2x2_telatar(snrdB,T,L,epsilon)

Compute the metaconverse upper bound [2, Eq. (41)] by invoking the function
MC USTM 2x2() and by optimizing over the input diagonal matrices {k }lk=1
and the number of effectively used transmit antennas. The optimization over
the {k }lk=1 is restricted to 2 2 diagonal matrices whose diagonal entries are
either [SNR/2, SNR/2] or [SNR, 0] according to the generalized form of Telatar
conjecture reported in [2, Eq. (42)]. NOTE: this function may be very slow if L
is large or epsilon is small.

3.4 Alamouti: DT USTM Alamouti(), MC USTM Alamouti()

Format

function [R,current_eps]
=MC_USTM_Alamouti(snrdB,T,L,epsilon,prec,filename)
function [R,current_eps]
=DT_USTM_Alamouti(snrdB,T,L,epsilon,prec,filename)

These functions compute achievability and converse bounds for a 2 2 MIMO

system with Alamouti used as inner code

3.5 Outage: outage mi(), outage alamouti(),

outage 2x2 telatar()
function [Cout,current_eps]
=outage_mi(snrdB,epsilon,Mt,Mr,L,pow_all,prec)
14 CHAPTER 3. BLOCK-MEMORYLESS RAYLEIGH FADING (NO CSI)

function [Cout,current_eps]
=outage_alamouti(snrdb,epsilon,L,prec,filename)
function R=outage_2x2_telatar(L,prec,epsilon,snrdb)

The first two functions compute the outage mutual information I for a given
set of diagonal input covariance matrices, and the outage mutual information
with Alamouti, respectively:

P[E[(X l ; Yl )|Hl ] I ] =  ,
where in the first case, the covariance structure of X l = (X1 , . . . , Xl ) is specified
by pow_all and
l
l l
X dPYk |Xk
(X ; Y ) = log .
dPYk
k=1

The outage_mi function evaluates the outage mutual information corresponding

to a given input covariance matrix, whose eigenvalues are specified in pow_all.
Specifically, pow_all is a Mt-1 x L matrix containing the first Mt1 eigenvalues
of the input covariance matrices corresponding to each coherence interval. The
last eigenvalue follows from the power constraint. To obtain the outage capacity
one has to optimize over all possible choices of pow_all.
The function outage_2x2_telatar provides an example on how to perform
this optimization for the 22 MIMO case. The optimization is restricted to input
covariance matrices that satisfy the generalized Telatar conjecture [2, Eq. (42)]
according to which in the eigenvalues of the input covariance matrix in each
coherence interval satisfy the following property: k out the Mt eigenvalues are
equal to SNR/k and the remaining Mtk eigenvalues are equal to zero. The
value of k may change across the coherence intervals.

3.6 Ergodic capacity: ergodic USTM()

function RergUSTM=ergodic_USTM(snrdB,T,Mt,Mr,prec)

This function computes a lower bound on the ergodic capacity (no CSIR),
obtained by evaluating the mutual information for a USTM input distribution.
It is assumed that mt mr .
Chapter 4

Quasi-static fading channel

General info:

Maintainer: W. Yang <[email protected]>

Main references: [3]

Example: See Fig. 6.1

Channel model: a quasi-static MIMO fading channel with mt transmit

and mr receive antennas. The channel input-output relation within n
channel uses is given by
Y = XH + Z (4.1)
where X Cnmt is the signal transmitted over n channel uses; Y Cnmr
is the corresponding received signal; the matrix H Cmt mr contains the
complex fading coefficients, which are random but remain constant over
the n channel uses; Z Cnmr denotes the additive noise at the receiver,
which is independent of H and has iid CN (0, 1) entries. Power constraints:
each codeword xn satisfies

kXk2F nP.

In the code, we also assume that the entries of H are iid Rician-distributed,
i.e., p
Hij CN ( K/(K + 1), 1/(K + 1))
where K is the Rician K-factor.

The code for the MIMO case (mt > 1) is built under the additional
assumption that all codewords X are isotropic, i.e., (see [3, Sec. III])

1 H P
X X= Im . (4.2)
n mt t
Currently, we do not have code for MIMO achievability and converse
bounds that hold without the assumption (4.2).

Input/output arguments:

1. nn input argument; vector of blocklengths.

15
16 CHAPTER 4. QUASI-STATIC FADING CHANNEL

MetaConverse
EpsilonCapacity

0.8

NormalApproximation
YLabel

0.6 KappaBetaBound

AchNocsi

0.4

0.2

0
0 100 200 300 400 500 600 700 800 900 1000
XLabel

Figure 4.1: Achievability and converse bounds for a quasi-static SIMO Rician-
fading channel with K-factor equal to 20 dB, two receive antennas, SNR = 1.55
dB, and  = 103 .

2. error input argument; block error probability.

3. tx, rx input arguments; number of transmit and receive antennas.
4. P input argument; parameter of the power constraint (linear scale).
5. K input argument; Rician K-factor. If K is not set, it defaults to 0,
in which case the entries of H follow a Rayleigh distribution.
6. rate_a, rate_c, rate_na output arguments; lower (achievabil-
ity) bound, upper (converse) bound, and normal approximation of
(log2 M )/n, channel coding rate (base-2 information units).

4.1 Summary plot: plot all()

Format:

[C_e, rate_a, rate_c, rate_na] = plot_all(nn, P, error, tx, rx, K);

This function plots capacity/achievability/converse/normal approximation

curves on the same figure. See Fig. 6.1.

4.2 Achievability: ach simo csir()

Format:

function rate_a = ach_simo_csir(nn, P, error, rx, K);

This function computes the achievability bound over a single-input

multiple-output (SIMO) channel with perfect CSIR, see [3, Footnote 7].
4.3. ACHIEVABILITY: ACH SIMO NOCSI() 17

4.3 Achievability: ach simo nocsi()

Format:
function rate_a = ach_simo_nocsi(nn, P, error, rx, K);
This function computes the achievability bound over a SIMO channel with
no CSI, see [3, Eq. (67)].

4.4 Achievability: mimo iso ach()

Format:
function rate_a = mimo_iso_ach(nn, P, error, tx, rx, K);
This function computes the achievability bound over a MIMO channel with
no CSI, where all input codewords satisfy XH X = nP
mt Imt , see [3, Eq. (65)].

4.5 Converse: converse simo()

function rate_c = converse_simo(nn,P, error, rx, K);
This function computes the converse bound over a single-input multiple-
output (SIMO) channel with CSIRT, see [3, Eq. (76)].

4.6 Converse: mimo iso conv()

function rate_c = mimo_iso_ach(nn, P, error, tx, rx, K);
This function computes the converse bound over a MIMO channel with
CSIRT, where all input codewords satisfy XH X = nP
mt Imt , see [3, Eq. (78)].

4.7 Normal approximation: normapprox simo(),

normapprox mimo iso()
Format:
function rate_na = normapprox_simo(nn,P, error, rx, K)
function rate_na = normapprox_mimo_iso(nn, P, error, tx, rx, K);
These two functions compute the normal approximation RN (, n) to both
achievability and converse, obtained by solving the following equation
" !#
C(H) RN (, n)
=E Q p
nV (H)
where  
P
C(H) , log2 det Imt + HHH
mt
and
m
X log22 e
V (H) , m log22 e
i=1
(1 + P i /mt )
with m , min{mt , mr } and {i } being the eigenvalues of HHH .
Chapter 5

Quasi-static fading channel

(long-term power constraint)

General info:
Maintainer: W. Yang <[email protected]>
Main references: [4]
Channel model:
Yi = HXi + Zi , i = 1, . . . , n (5.1)
where
Xi , Yi C are the channel input and output, respectively
H C is the complex channel coefficient
Zi CN (0, 1) is the i.i.d. white Gaussian noise
An (n, M, )lt code for the channel (5.1) consists of
an encoder f : {1, . . . , M } C Cn that maps the message
W {1, . . . , M } and the channel coefficient H to a codeword xn =
f (W, H) satisfying the long-term (i.e., averaged-over-all-codeword)
power constraint
EW,H kf (W, H)k2 nP
 
(5.2)
where W is equiprobable on {1, . . . , M }
a decoder g : Cn C {1, . . . , M } satisfying the average error
probability constraint
P[g(Y n , H) 6= W ]  (5.3)
where Y n is the channel output induced by the transmitted codeword
xn = f (W, H) according to (5.1).
Remark: for comparison, a code is said to satisfy a short-term (i.e.,
per-codeword) power constraint if the encoder satisfies
kf (W, H)k2 nP (5.4)
with probability one, see Chapter 4.

19
 CHAPTER 5. QUASI-STATIC FADING CHANNEL (LONG-TERM
20 POWER CONSTRAINT)
Common input/output arguments:
snr_db input argument; SNR in dB
nn input argument; blocklength
error input argument; block error probability
rate_ach_lt/rate_conv_lt/rate_na_lt output arguments; lower
(achievability) bound, upper (converse) bound, and normal approxi-
mation of (log2 M )/n, maximal achievable rate (in bits/ch. use).

5.1 The AWGN channel (H = 1)

Format

function rate_ach_lt=awgn_ach_lt(snr_db, nn, error);

function rate_conv_lt=awgn_conv_lt(snr_db, nn, error);
function rate_na_lt = awgn_na_lt(snr_db,nn,error);

These functions compute achievability/converse/normal approximation of

(log2 M )/n, maximal achievable rate (in bits/ch. use). The bounds are from [4,
Section II.C] .

5.2 Quasi-static fading channel with perfect CSIT

Under construction.
Chapter 6

Block-memoryless Rayleigh
Fading (CSIR)

General info:
Maintainer: A. Collins <[email protected]>
Main references: Coming Soon
Example: See Figure 6.1
Channel Model: with nt transmit antennas, nr receive antennas, and
coherence time T , the model is given by
Yj = Hj Xj + Zj , j = 1, . . . , n
Where F is either R or C, and
Xj Fnt T is channel input
Hj Fnr nt is fading matrix, iid Rayleigh per block
Zj Fnr T is the i.i.d. (circularly symmetric) Gassian noise matrix.
Yj Fnr T is output matrix
The input must satisfy the per-codeword power constraint
n
X
kxj k2F nP
j=1

Input / output arguments

P Power constraint
n number of coherent blocks (for nT total channel uses)
n_t number of transmitter antennas
n_r number of receiver antennas
T channel coherence time (in number of channel uses)
epsilon average probability of error
real_or_complex specifies real or complex channel (set to real
or complex, respectively)

21
22 CHAPTER 6. BLOCK-MEMORYLESS RAYLEIGH FADING (CSIR)

6.1 Summary plot: plot all()

Format:
function log_M = plot_all(P,n_t,n_r,T,epsilon)
This will plot the achievability bound, the normal approximation, and the
capacity as in Figure 6.1. To change the plot range, the n_vec vector can be
changed to include the desired points.

Figure 6.1: Achievability (blue), normal approximation (red), and capacity for
the real MIMO CSIR block fading channel with nt = nr = T = 2, P = 10,
 = 103 .

6.2 Achievability Bound: MIMO achievability CSIR()

Format:
function log_M = MIMO_achievability_CSIR(n,epsilon,n_t,n_r,T,P,real_or_complex)
This function computes the following average probability of error achievabil-
ity bound:

(PY , QY )
log M sup sup
0< < QY 1+ (PXY , PX QY )
Where PY is the output distribution through the channel when the input PX
has the following distribution (X n iid N (0, P/nt ))

nT P
Xn
kX n kF
6.3. NORMAL APPROXIMATION: MIMO CSIR NORMAPX() 23

And QY is the output distribution when the input is iid Gaussian N (0, P/nt ).
Both of these are fairly easily adjustable in the code.
Note that the iter variable in the code sets the number of Monte Carlo
iterations. The sampling loop uses parfor, which attempts to distribute the
computation. The number of Monte Carlo iterations should be taken < 100/.
For small  (e.g. 103 , you may want to let the code run overnight.

6.3 Normal Approximation: MIMO csir normapx()

Format:

function logM = MIMO_csir_normapx(n,epsilon,n_t,n_r,T,P,real_or_complex)

This function computes the normal approximation up to the O( n) term,
i.e.

log M nT C nT V Q1 ()

It utilizes one of two functions to compute the capacity C and dispersion V :

function [C,V] = capacity_and_dispersion(n_t,n_r,T,P)

or

function [C,V] = capacity_and_dispersion_mc(n_t,n_r,T,P,real_or_complex)

The capacity_and_dispersion function computes C and V via numerical inte-

gration for the complex MIMO channel, while capacity_and_dispersion_mc
compute these quantities using Monte Carlo. The Monte Carlo computation is
also useful its flexibility, e.g. its simple to change the fading distribution to
compute the capacity and dispersion of non-Rayleigh fading channels.
Chapter 7

Energy-per-bit under vanishing

spectral efficiency

General info:

Maintainer: Y. Polyanskiy <[email protected]>

Main references: [5, 6]

Example: See Fig. 7.1

Channel model:

Yj = Hj Xj + Zj , j = 1, . . . n= (!!!) (7.1)

Noise: Zj CN (0, N0 ) iid and N0 is noise spectral density

Channel gain (for AWGN): Hj = 1 constant
Channel gain (for Rayleigh fading): Hj CN (0, 1) iid
Power constraints: Each codeword x satisfies

X
|xj |2 E ,
j=1

where E is the total energy budget.

Energy-per-bit: For the codebook C of size M its energy-per-bit is
defined as
E
Eb = .
log2 M
Description:
Bounds in this Chapter address the question of finding the absolutely
minimal energy-per-bit required for transmitting log2 M bits with prob-
ability of error  and without restriction on the number of channel uses.
Asymptotically as M it is known:

Eb
loge 2 1.59 dB
N0

25
 CHAPTER 7. ENERGY-PER-BIT UNDER VANISHING SPECTRAL
26 EFFICIENCY
> energy-per-bit/plot_all()
Energy per bit vs. length of the message (Rayleigh fading); =0.001
Energy per bit vs. length of the message (AWGN channel); =0.001
8 10
Achievability noCSI: Achievability (ach_ht)
Normal approximation
Converse
noCSI: Normal approximation
7
Shannon limit, 1.59 dB noCSI: Converse
8 CSIR: Normal approximation
6
Shannon limit, 1.59 dB

5
6

Eb/No, dB
4
Eb/No, dB

4
3

2
2
1

0 0

2 0 1 2 3 4 5
2 0 1 2 3 4 5 6 10 10 10 10 10 10
10 10 10 10 10 10 10
Information bits, k Information bits, k

Figure 7.1: Example figures generated by the energy-per-bit code

Real/Complex: In the regime of infinite bandwidth as here, there is no

difference between real/complex channel inputs (but noise variance in the
real case should be taken Zj N (0, N20 )).
MIMO and block-fading: Surprisingly, having multiple transmit antennas
or block-constant fading makes no difference for energy-per-bit, see [6].
N0
Having mr > 1 receive antennas is equivalent to scaling N0 m r
and
thus again does not require any special treatment.
Remark: These bounds correspond to allowing no feedback link. Even a
modicum of feedback significantly improves the energy-per-bit tradeoff at
finite length, see [5].
Eb
Remark: The tradeoff N 0
vs. log2 M can also be interpreted as the
rate-blocklength tradeoff for the power-constrained infinite-bandwidth
continunuous-time channel, see [5, (50)-(51)].
Common input/output arguments:
1. epsil block probability of block error.
2. lm or Lms input or output argument; log2 M , log-size of codebook
(base-2 information units).
3. en0 or EE input/output argument; total energy budget available
Eb
EE= NE0 (to get N0
divide by lm).

7.1 Summary plot

plot_all()

This script generates two figures, shown on Fig. 7.1.

Note: Script takes several hours to finish.

7.2 The AWGN channel (Hj = 1)

Here Hj = 1 in the channel model (7.1).
Format:
7.3. RAYLEIGH FADING: CSIR CASE 27

function En0 = energy_awgn_ach(Lms, epsil);

function lm = energy_awgn_conv(en0, epsil);
function lm = energy_awgn_normapx(en0, epsil);

These functions return achievability/converse/normal approximation. The

bounds are from [5, Theorems 2 and 3].

7.3 Rayleigh fading: CSIR case

Format:

function lm = energy_csir_conv(EE, epsil);

This function computes the converse (upper) bound on the number of

information bits, lm, that a Rayleigh fading channel with perfect CSIR can
transmit under a total energy budget EE= NE0 ; see [6, Theorem 11]. Note that
for CSIR case we can use energy_aegn_ach() and energy_awgn_normapx()
for achievability and normal approximation, see [6, Section III.D].
Note: vector EE should be in ascending order.

7.4 Rayleigh fading: noCSI case

Format:

function lm = energy_nocsi_ach_ht(EE, epsil);

function lm = energy_nocsi_ach_ml(EE, epsil);
These are two achievability bounds [6, Corollary 4 and 5] with the second
one, _ml(), being typically better.

function lm = energy_nocsi_conv(EE, epsil);

function lm = energy_nocsi_normapx(EE, epsil);

The converse bound is [6, Theorem 6]. The normal approximation is com-
puted by maximizing over A in [6, (64)].
Note: Vector EE should be in ascending order.
Note: Functions energy_nocsi_ach_ml() and energy_nocsi_conv() are
very slow.
Chapter 8

Fixed-length source coding under

a distortion criterion

Maintainer: V. Kostina <[email protected]>

Main references: [7]
Performance criterion - excess distortion probability:

P[d(S, Z) > d] ,

where S is the source and Z is its representation.

The code uses auxiliary functions contained in lib/. This directory
can to be added to Matlab path by running the following command:
path(path, ../../lib);
To speed up calculations, some functions use the so-called persistent
variables whose values are retained during multiple calls to those functions.
This helps to choose the starting positions for various optimizations
wisely. For example, if the function is computed for blocklength 510 after
blocklength 500, it takes advantage of the previous computation to choose
a starting position for various optimizations. To avoid problems, use
command clear all to clear the values of all persistent variables when
making a computation unrelated to the previous computation.
Recommended usage:

path(path, ../../lib);
clear all;

N = 10:10:1000;
out = NaN(1, length(N));
for k = 1:length(N)
n = N(k);
try
out(k) = Rstar(..., n, ...);
save(mydatafile);
catch ME

29
 CHAPTER 8. FIXED-LENGTH SOURCE CODING UNDER A
30 DISTORTION CRITERION
> cd sc/GMS
> plot_all(0.1, 0.01, 10:5:500)
Lossy compression of iid standard normal source. Distortion=0.1 at excess=1%
3.6
Rogers (achievability)
Normal approximation
3.4 Sphere covering + HT (converse)

3.2

Compression rate, bit / source sample

3

2.8

2.6

2.4

2.2

1.8

1.6
0 50 100 150 200 250 300 350 400 450 500
Blocklength, n

Figure 8.1: Lossy compression bounds for the memoryless Gaussian source.

fprintf(Error: %s\n, getReport(ME));

out(k) = NaN;
end
end

For an example plot (produced via sc/GMS/plot_all.m) see Fig. 8.1.

8.1 Binary memoryless source

Source model
S1 , S2 , . . . are i.i.d Bernoulli with bias p.

Distortion measure
Bit error rate
k
1X
d(sk , z k ) = 1{si 6= zi }.
k i=1

Matlab code
sc/BMS/

Main functions
Rstar(d, n, e, p, fun): Compression rate compatible with a given excess
distortion.

Dstar(R, n, e, p, fun): Distortion threshold compatible with a given excess

probability and a given rate.
8.2. GAUSSIAN MEMORYLESS SOURCE 31

Input arguments
d: distortion threshold.

R: compression rate.

n: blocklength.

e: excess distortion probability.

p: source bias.

fun: alias for the function to be computed. Acceptable values:

shannon : Shannons achievability bound [7, Theorem 1].

normal : Gaussian approximation [7, Theorem 12].
spherecoveringc : Converse via sphere covering and hypothesis testing [7,
Theorem 8].
generalc : General converse via d-tilted information [7, Theorem 7].
spherecoveringa : Achievability via covering with random spheres [7,
Theorem 21].
permutationa : Achievability via permutation coding (i.e. all code-
words are of the same type, namely, the type that achieves the
rate-distortion function) [7, Theorem 22].
martona : Converse via Martons bound [7, Theorem 5].

8.2 Gaussian memoryless source

Source model
S1 , S2 , . . . are i.i.d N (0, 1).

Distortion measure
Squared error
k
1X
d(sk , z k ) = (si zi )2
k i=1

Matlab code
sc/GMS/

Main functions
Rstar(d, n, e, fun): Compression rate compatible with a given excess dis-
tortion.

Dstar(R, n, e, fun): Distortion threshold compatible with a given excess

probability and a given rate.
 CHAPTER 8. FIXED-LENGTH SOURCE CODING UNDER A
32 DISTORTION CRITERION
Input arguments
d: distortion threshold.

R: compression rate.

n: blocklength.

e: excess distortion probability.

fun: alias for the function to be computed. Acceptable values:

shannon : Shannons achievability bound [7, Theorem 1].

normal : Gaussian approximation [7, Theorem 12].
spherecoveringc : Converse via sphere covering and hypothesis testing [7,
Theorem 8].
spherecoveringa : Achievability via covering with random spheres [7,
Theorem 37].

The following fun values are accepted only by Rstar(...).

generalc : General converse via d-tilted information [7, Theorem 7].

rogersa : Achievability using a result by Rogers [7, Theorem 39].

8.3 Binary erased source

Source model
S1 , S2 , . . . are i.i.d fair coin flips. However, the encoder does not observe
S1 , S2 , . . .. Instead, it observes an erased version of the source: X1 , X2 , . . .,
where Xi = Si with probability 1 , and Xi = ? with probability .

Distortion measure
Bit error rate with respect to the original source

k
1X
d(sk , z k ) = 1{si 6= zi }, si , zi {0, 1}.
k i=1

Matlab code
sc/BES/

Main functions
Rstar(d, n, e, alpha, fun): Compression rate compatible with a given ex-
cess distortion.
8.3. BINARY ERASED SOURCE 33

Input arguments
d: distortion threshold.
n: blocklength.
e: excess distortion probability.
alpha: erasure probability.
fun: alias for the function to be computed. Acceptable values:
spherecoveringa : Achievability via covering with random spheres [7,
Theorem 33].
spherecoveringc : Converse via sphere covering (hypothesis testing) [7,
Theorem 32].
normal : Gaussian approximation [7, Theorem 12].
aeq : Achievability via covering with random spheres for a surrogate
rate-distortion problem which negates the stochastic variability of
the erasure channel [8].
ceq : Converse via sphere covering (hypothesis testing) for the surrogate
problem [8].
normaleq : Gaussian approximation for the surrogate problem [8].
Chapter 9

Variable-length compression
allowing errors

Maintainer: V. Kostina <[email protected]>

Main references: [9].
Fixed probability of error:

P[S 6= Z] ,

where S is the source and Z is its reconstruction after the (almost lossless)
encoding.
Performance criterion: average length of the compressed representation.
Performance criterion - excess distortion probability:

P[d(S, Z) > d] ,

where S is the source and Z is its reconstruction after the transmission

over a noisy channel.
The code uses auxiliary functions contained in lib/. This directory
needs to be added to Matlab path by running the following command:
path(path, ../../lib);
To speed up calculations, some functions use the so-called persistent
variables whose values are retained during multiple calls to those functions.
This helps to choose the starting positions for various optimizations
wisely. For example, if the function is computed for blocklength 510 after
blocklength 500, it takes advantage of the previous computation to choose
a starting position for various optimizations. To avoid problems, use
command clear all to clear the values of all persistent variables when
making a computation unrelated to the previous computation.
Recommended usage:

path(path, ../../lib);
clear all;

35
 CHAPTER 9. VARIABLE-LENGTH COMPRESSION ALLOWING
36 ERRORS

N = 10:10:1000;
out = NaN(1, length(N));
for k = 1:length(N)
n = N(k);
try
out(k) = Rstar(..., n, ...);
save(mydatafile);
catch ME
fprintf(Error: %s\n, getReport(ME));
out(k) = NaN;
end
end

9.1 Binary memoryless source

Source model
S1 , S2 , . . . are i.i.d Bernoulli with bias p.

Matlab code
sc/varrate/

Main function
Rstar(p, e, n, fun): Compression rate compatible with a given excess dis-
tortion.

Input arguments
p: source bias.
e: error probability.
n: blocklength.
fun: alias for the function to be computed. Acceptable values:
approx : Gaussian approximation [9, Theorem 4].
exact : Exact minimum average rate [9, (26)].
exacti : Exact value of the Erokhin function [9, (40)].
ie : Exact value of the expectation of the -cutoff of information in
S k [9, (11),(13)].
Chapter 10

Joint source-channel coding

under a distortion criterion

Maintainer: V. Kostina <[email protected]>

Main references: [10]
Performance criterion - excess distortion probability:

P[d(S, Z) > d] ,

where S is the source and Z is its reconstruction after the transmission

over a noisy channel.
The code uses auxiliary functions contained in lib/. This directory
needs to be added to Matlab path by running the following command:
path(path, ../../lib);
To speed up calculations, some functions use the so-called persistent
variables whose values are retained during multiple calls to those functions.
This helps to choose the starting positions for various optimizations
wisely. For example, if the function is computed for blocklength 510 after
blocklength 500, it takes advantage of the previous computation to choose
a starting position for various optimizations. To avoid problems, use
command clear all to clear the values of all persistent variables when
making a computation unrelated to the previous computation.
Recommended usage:

path(path, ../../lib);
clear all;

N = 10:10:1000;
out = NaN(1, length(N));
for k = 1:length(N)
n = N(k);
try
out(k) = Rstar(..., n, ...);
save(mydatafile);

37
 CHAPTER 10. JOINT SOURCE-CHANNEL CODING UNDER A
38 DISTORTION CRITERION
catch ME
fprintf(Error: %s\n, getReport(ME));
out(k) = NaN;
end
end

10.1 Binary memoryless source transmitted over a

binary symmetric channel
Source model
S1 , S2 , . . . are i.i.d Bernoulli with bias p.

Channel model
Yi = Xi + Ni , where Ni are i.i.d. Bernoulli with bias .

Distortion measure
Bit error rate
k
1X
d(sk , z k ) = 1{si 6= zi }.
k i=1

Matlab code
jscc/BMS-BSC/

Main functions
Rstar(p, delta, d, e, n, fun): Compression rate compatible with a given
excess distortion.

Dstar(R, n, e, p, delta, fun): Distortion threshold compatible with a

given excess probability and a given rate.

Input arguments
p: source bias.

delta: channel crossover probability.

d: distortion threshold.

e: excess distortion probability.

R: compression rate.

n: blocklength.

fun: alias for the function to be computed. Acceptable values:

approx : Gaussian approximation [10, Theorem 10].

 39
10.2. GAUSSIAN SOURCE TRANSMITTED OVER AN AWGN CHANNEL

clistdecoding : Converse via list decoding and hypothesis testing [10,

Theorem 4].
ajoint : Achievability via joint source-channel coding [10, Theorem 8].
aseparate : Achievability via separation [10, Theorem 6].

The following fun values are accepted only by Rstar(...).

cgeneral : General converse via d-tilted information [10, Theorem 2].

approxseparate : Gaussian approximation for the separation architecture
[10, (126)].
arcu : For the almost lossless transmission, achievability via RCU
bound [10, Theorem 9].

The following fun values are accepted only by Dstar(...).

random : Achievability via random coding for fair coin flip source [10,
Theorem 6].
nocoding : Achievability via no coding. [10, Theorem 23].

10.2 Gaussian source transmitted over an AWGN

channel
Source model
S1 , S2 , . . . are i.i.d N (0, 1).

Channel model
Yi = Xi + Ni , where Ni are i.i.d. N (0, 1).

Distortion measure
Squared error
k
k k 1X
d(s , z ) = (si zi )2 .
k i=1

Matlab code
jscc/GMS-AWGN/

Main functions
Rstar(P, d, e, n, fun): Transmission rate compatible with a given excess
distortion (nats).

Dstar(R, n, e, P, fun): Distortion threshold compatible with a given excess

probability and a given rate.
 CHAPTER 10. JOINT SOURCE-CHANNEL CODING UNDER A
40 DISTORTION CRITERION
Input arguments
P: maximum allowable transmit power, or SNR.
d: distortion threshold.
e: excess distortion probability.
R: transmission rate.
n: blocklength.
fun: alias for the function to be computed. Acceptable values:

approx : Gaussian approximation [10, Theorem 10].

cht : Converse via list decoding and hypothesis testing [10, Theorem 4].
ajoint : Achievability via joint source-channel coding [10, Theorem 8].
aseparate : Achievability via separation [10, Theorem 6].

The following fun values are accepted only by Dstar().

nocoding : Achievability via no coding [10, Theorem 24].
approxnocoding : Gaussian approximation for the uncoded system [10,
(129), (208)].
The following fun values are accepted only by Rstar().
cgeneral : General converse via d-tilted information [10, Theorem 2].
approxseparate : Gaussian approximation for the separation architecture
[10, (126)].
Chapter 11

The binary-symmetric channel

(BSC)

General info:
Maintainer: Y. Polyanskiy <[email protected]>
Main references: [1]
Example: See Fig. 11.1
Channel model:

Yj = Xj + Zj mod 2, j = 1, . . . , n

Codewords are 0/1: xj {0, 1}, j = 1, . . . , n

Noise: Zj Ber() iid.

Common input/output arguments:

1. delta input argument; crossover probability .
2. epsil block probability of block error. Warning: All code in this
section is tailored to maximal probability of error.
3. lm or Lms output argument; log2 M , log-size of codebook (base-2
information units).
4. n input argument; blocklength.
For slow functions that do not support vectorized arguments.
5. Ns input argument; vector of blocklengths.

11.1 Summary plot: plot all()

Definition:
function [Ns ach_rcu ach_dt ach_gal conv normapx_val] = plot_all(delta, epsil);
This function computes and plots various bounds on the same figure. See
Fig. 11.1.

41
42 CHAPTER 11. THE BINARY-SYMMETRIC CHANNEL (BSC)

> plot_all()

Bounds for the BSC(0.11), Pe,max = 0.001

0.5

0.45

0.4

0.35

0.3
Rate, R

0.25

0.2

0.15

Capacity
0.1
Converse
Normal approximation
0.05 RCU achievability
DT achievability
Gallager achievability
0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Blocklen, n

Figure 11.1: Code and resulting picture for BSC bounds

11.2 Achievability: dt ach()

Definition:

function Lms = dt_ach(Ns, delta, epsil)

Function computes DT achievability bound, see [1, (166)].

11.3 Achievability: rcu ach()

Format:

function lm = rcu_ach(n, delta, epsil)

This computes the RCU achievability bound, see [1, (162)].

Warning: This function only takes scalar value of n.

11.4 Achievability: gallager ach()

Format:

function Lms = gallager_ach(Ns, delta, epsil)

Computes Gallagers achievability bound, see [1, (44)].

11.5 Converse: converse()

Format:
11.6. NORMAL APPROXIMATION: NORMAPX() 43

function Lms = converse(Ns, delta, epsil)

This computes the sphere-packing lower bound. Equivalently, this is the

best possible meta-converse bound with QY n = Ber(1/2)n , see [1, Theorem 35].

11.6 Normal approximation: normapx()

Format:

function Lms = normapx(Ns, delta, epsil);

Fast and frequently very precise approximation to both achievability and
converse:
1
log M nC nV Q1 () + log n
2
where C = log 2 log 1
(1 ) log 1
1 and V = (1 ) log2 1
.
Chapter 12

The binary-erasure channel

(BEC)

General info:

Maintainer: Y. Polyanskiy <[email protected]>

Main references: [1]

Example: See Fig. 11.1

Channel model:
(
Xj , with prob.1
Yj = j = 1, . . . , n
e, with prob.

Codewords are 0/1: xj {0, 1}, j = 1, . . . , n

Common input/output arguments:

1. delta input argument; crossover probability .

2. epsil block probability of block error. Warning: All code in this
section is tailored to maximal probability of error.
3. lm or Lms output argument; log2 M , log-size of codebook (base-2
information units).
4. n input argument; blocklength.
For slow functions that do not support vectorized arguments.
5. Ns input argument; vector of blocklengths.

12.1 Summary plot: plot all()

Definition:

function [Ns ach_dt ach_gal conv normapx_val] = plot_all(delta, epsil);

This function computes and plots various bounds on the same figure. See
Fig. 12.1.

45
46 CHAPTER 12. THE BINARY-ERASURE CHANNEL (BEC)

> plot_all()

Bounds for the BSC(0.5), Pe,max = 0.001

0.5

0.45

0.4

0.35

0.3
Rate, R

0.25

0.2

0.15

0.1
Capacity
Converse
0.05 Normal approximation
DT achievability
Gallager achievability
0
0 100 200 300 400 500 600 700 800 900 1000
Blocklen, n

Figure 12.1: Code and resulting picture for BEC bounds

12.2 Achievability: dt ach()

Definition:

function Lms = dt_ach(Ns, delta, epsil)

Function computes DT achievability bound, see [1, (183)].

Note: The RCU bound is weaker for the BEC, so we do not need to evaluate
it.

12.3 Achievability: gallager ach()

Format:

function Lms = gallager_ach(Ns, delta, epsil)

Computes Gallagers achievability bound, see [1, (44)].

12.4 Converse: converse()

Format:

function Lms = converse(Ns, delta, epsil)

This computes the special converse bound for the BEC, see [1, Theorem
38]. Later, it was found that it corresponds to the meta-converse bound with a
non-product distribution QY n , see [11, Section VI.D].
12.5. NORMAL APPROXIMATION: NORMAPX() 47

12.5 Normal approximation: normapx()

Format:

function Lms = normapx(Ns, delta, epsil);

Fast and frequently very precise approximation to both achievability and
converse:
log M nC nV Q1 () ,
where C = (1 ) log 2 and V = (1 ) log2 2.
Chapter 13

The binary-AWGN channel

(BI-AWGN)

General info:
Maintainer: T. Erseghe <[email protected]>
Main references: [12]
Example: See Fig. 13.1
Channel model:

Yj = Xj + Zj , j = 1, . . . , n

Codewords are 0/1: xj {1}, j = 1, . . . , n

Noise: Zj N (0, 1) iid
is the signal to noise ratio (SNR).

Common input/output arguments:

1. n blocklength.
2. Pe block probability of block error, or packet error rate. Warning:
All code in this section is tailored to average probability of error.
3. R code rate.
4. SNRdB signal to noise ratio in dB.

13.1 Converse: converse mc()

Definition #1:
R = converse_mc(N,Pe,SNRdB,approx)
returns the metaconverse Polyanskiy-Poor-Verdu (PPV) upper bound on rate
for a codeword of length N, a packet error rate of Pe, and a signal to noise
ratio of SNRdB (in dB). Multiple bound values can be obtained by ensuring that
N, Pe, and SNRdB have the same length. The specific approximation used for
calculating the bound is specified by approx:

49
50 CHAPTER 13. THE BINARY-AWGN CHANNEL (BI-AWGN)

normal normal approximation, the O(n1 ) approximation available

from [12, Fig. 6, 3rd block]; this is a very fast algorithm, but may provide
a weak approximation.

On2 (default) the O(n2 ) approximation derived from [12, Fig. 6, 2nd
block]; at low Pe this is much more reliable than the normal option.

On3 the O(n3 ) approximation derived from [12, Fig. 6, 2nd block];
at high Pe this is much more unrealiable than On2. It can be used to
identify the region where On2 is a reliable solution, namely the region
where On2 and On3 closely match.

full (this option will be available soon) the full bound derived numer-
ically from [12, Fig. 6, 1st block].

Definition #2:

SNRdB = converse_mc(N,Pe,R,approx,error)

for a codeword of length N, and code rate R, the function returns the signal to
noise ration SNRdB (in dB) at which the PPV metaconverse lower bound on
error rate is equal to Pe. Multiple bound values can be obtained by ensuring
that N, Pe, and R have the same length. The specific approximation used for
calculating the bound is specified by approx (see above).

13.2 Example 1: example speff.m

In example_speff.m we use the converse_mc function in order to derive an
upper bound on rate, R, and then plot a bound on spectral efficiency, namely
= 2R. We also recall the relation

Eb
=
N0
with the SNR, and Eb the energy spent per bit of information. The example
is built on a codeword of length n=1000 for which the bounds at Pe=1e-5 are
derived according to the following code:

n = 1e3;
Pe = 1e-5;
SNRdB = -15:10; % SNR Omega
uno = ones(size(SNRdB)); % vector of ones

% normal approximation - expected execution time 2 sec

rhoNA = 2*converse_mc(n*uno,Pe*uno,SNRdB,normal);
EbN0NA = SNRdB-10*log10(rhoNA);

% O(n^-2) approximation - expected execution time 90 sec

rhoPPV = 2*converse_mc(n*uno,Pe*uno,SNRdB,On2);
EbN0PPV = SNRdB-10*log10(rhoPPV);
13.3. EXAMPLE 2: EXAMPLE PER.M 51

% O(n^-3) approximation - expected execution time 120 sec

rhoPPVb = 2*converse_mc(n*uno,Pe*uno,SNRdB,On3);
EbN0PPVb = SNRdB-10*log10(rhoPPVb);

and then the results is plotted by using

figure
set(0,defaulttextinterpreter,latex)
semilogy(EbN0PPV,rhoPPV,-,EbN0PPVb,rhoPPVb,--,...
EbN0NA,rhoNA,-.)
xlabel(SNR $E_b/N_0$)
ylabel(spectral efficiency $\rho$ [bit/s/Hz])
grid

The result is given in the plot on the left in Fig. 13.1

n = 1000, Pe = 1e - 05 n = 1000, R = 0. 5
1 0
10 10

1
10
sp e c t r al e ffi c i e n c y [ b i t /s/H z ]

0
10
2
10
O ( n 2 ) ap p r ox
e r r or p r ob ab i l i ty P e

O ( n 3 ) ap p r ox
3
10
n or m al ap p r ox
1
10
4
10

5
10
2
10

6
10

3 7
10 10
0 5 10 15 0.5 0 0.5 1 1.5 2
S N R E b /N 0 S N R E b /N 0

Figure 13.1: Examples of BI-AWGN metaconverse bounds

13.3 Example 2: example per.m

In example_per.m we use the converse_mc function in order to derive a lower
bound on the packet error rate, Pe . The example is built on a codeword of
length n=1000 and rate R=0.5 for which bounds are derived according to the
following code:
52 CHAPTER 13. THE BINARY-AWGN CHANNEL (BI-AWGN)

n = 1e3;
R = 1/2;
Pev = 10.^([-0.5:-0.25:-1.75,-2:-0.5:-7]);
uno = ones(size(Pev));

% normal approximation - expected execution time 15 sec

SNRdBNA = converse_mc(n*uno,Pev,R*uno,normal,error);

% O(n^-2) approximation - expected execution time 120 sec

SNRdBPPV = converse_mc(n*uno,Pev,R*uno,On2,error);

% O(n^-3) approximation - expected execution time 280 sec

SNRdBPPVb = converse_mc(n*uno,Pev,R*uno,On3,error);

and then the results is plotted by using

figure
set(0,defaulttextinterpreter,latex)
semilogy(SNRdBPPV,Pev,-,SNRdBPPVb,Pev,--,...
SNRdBNA,Pev,-.)
xlabel(SNR $E_b/N_0$)
ylabel(error probability $P_e$)
grid
The result is given in the plot on the right in Fig. 13.1
Bibliography

[1] Y. Polyanskiy, H. V. Poor, and S. Verdu, Channel coding rate in the

finite blocklength regime, IEEE Trans. Inf. Theory, vol. 56, no. 5, pp.
23072359, May 2010.
[2] G. Durisi, T. Koch, J. Ostman, Y. Polyanskiy, and W. Yang, Short-
packet communications over multiple-antenna Rayleigh-fading channels,
IEEEJ C OM, vol. 64, no. 2, pp.618 629, F eb.2016.
[3] W. Yang, G. Durisi, T. Koch, and Y. Polyanskiy, Quasi-static multiple-
antenna fading channels at finite blocklength, IEEE Trans. Inf. Theory,
vol. 60, no. 7, pp. 42324265, Jul. 2014.
[4] W. Yang, G. Caire, G. Durisi, and Y. Polyanskiy, Optimum power control
at finite blocklength, IEEE Trans. Inf. Theory, vol. 61, no. 9, pp. 4598
4615, Sep. 2015.
[5] Y. Polyanskiy, H. V. Poor, and S. Verdu, Minimum energy to send k bits
with and without feedback, IEEE Trans. Inf. Theory, vol. 57, no. 8, pp.
48804902, Aug. 2011.
[6] W. Yang, G. Durisi, and Y. Polyanskiy, Minimum energy to send k bits
over multiple-antenna fading channels, arXiv:1507.03843v1, Jul. 2015.
[7] V. Kostina and S. Verdu, Fixed-length lossy compression in the finite
blocklength regime, IEEE Transactions on Information Theory, vol. 58,
no. 6, pp. 33093338, June 2012.
[8] , Nonasymptotic noisy lossy source coding, in Proceedings 2013
IEEE Information Theory Workshop, Seville, Spain, Sep. 2013.
[9] V. Kostina, Y. Polyanskiy, and S. Verdu, Variable-length compression
allowing errors, IEEE Transactions on Information Theory, vol. 61, no. 9,
pp. 43164330, Aug. 2015.
[10] V. Kostina and S. Verdu, Lossy joint source-channel coding in the finite
blocklength regime, IEEE Transactions on Information Theory, vol. 59,
no. 5, pp. 25452575, May 2013.
[11] Y. Polyanskiy, Saddle point in the minimax converse for channel coding,
IEEE Trans. Inf. Theory, vol. 59, no. 7, pp. 25762595, Jul. 2013.
[12] T. Erseghe, Coding in the finite-blocklength regime: Bounds based on
laplace integrals and their asymptotic approximations, arXiv preprint
arXiv:1511.04629, 2015.

53