Combinatorics in Space

The Mariner 9 Telemetry System

 Mariner 9 Mission

Launched: May 30, 1971

Arrived: Nov. 14, 1971
Turned Off: Oct. 27, 1972

Mission Objectives:
(Mariner 8): Map 70% of Martian surface.
(Mariner 9): Study temporal changes in Martian atmosphere and
surface features.
 Live TV

A black and white TV camera was used to broadcast “live”

pictures of the Martian surface.

Each photo-receptor in the camera measures the brightness of a

section of the Martian surface about 4-5 km square, and outputs a
grayness value in the range 0-63. This value is represented as a
binary 6-tuple.

The TV image is thus digitalized by the photo-receptor bank and is

output as a stream of thousands of binary 6-tuples.
 The Problem

The problem that arises comes from the difficulties inherent with
the transmission of messages. More precisely, we wish to transmit
a message and know that in the process of transmission there will
be some altering of the message, due to weak signals, sporadic
electrical bursts and other naturally occurring noise that creeps into
the transmission medium. The problem is to insure that the
intended message (our original transmission) is obtainable from
whatever is actually received.
 The Repeat Code
One simple approach to this problem is what is called a repeat code.
For instance, if we wanted to send the message BAD NEWS, we
could repeat each letter a certain number of times and send,
say,
BBBBBAAAAADDDDD NNNNNEEEEEWWWWWSSSSS.

Even if a number of these letters got garbled in transmission, the

intended message could be recovered from a received message that
might look like
BBBEBFAAAADGDDD . MNNNTEEEEEWWWSWRRSSS,

by a process called majority decoding, which in this case would

mean that for each block of 5 letters the intended letter is the one
which appears most frequently in the block.
 Probability
The problem with this approach is economical, the repeat code is not
very efficient. The increased length of the transmitted code, and thus the
increased time and energy required to transmit it, is necessary in order to
be able to decode the message properly, but how efficiently a coding
procedure uses this increase depends upon the coding scheme. Suppose,
in our example, that the probability that a letter is garbled in
transmission is p = 0.05 and so q = 1 - p = 0.95 is the probability that a
letter is correctly received. Without any coding, the probability of our 8
letter (spaces included) message being correctly received is
q8 = (0.95)8 = 0.66.

(In this calculation we are assuming that the error in transmitting a single
symbol is independent of which position the symbol is in. This is a
common simplifying assumption ... which may not be appropriate in real
world situations.)
 Probability
Using the repeat code, the probability of correctly decoding a given letter
from a block of 5 symbols is
q5 + 5q4p + 10q3p2
since there are three ways to decode correctly: 1) all the symbols are
correct, 2) one symbol is incorrect (5 ways this can happen) or 3) two
symbols are incorrect (10 ways this can happen) [notice that these are
just terms in the expansion of (q+p)5]. So we obtain
(.95)5 + 5(.95)4(.05) + 10(.95)3(.05)2 = 0.9988
and thus the probability of getting the correct eight letter message after
decoding is (0.9988)8 = 0.990, clearly a great increase over the non-
coded message ( = 0.66), but this 1% probability of getting the wrong
message might not be acceptable for certain applications.
 Terminology
To increase the probability of decoding the correct message with
this type of code we would have to increase the number of repeats
- a fix which may not be desirable or even possible in certain
situations. However, as we shall see, other coding schemes could
increase the probability to 0.9999 without increasing the length of
the coded message.

Before leaving the repeat codes to look at other coding

schemes, let us introduce some terminology. Each block of
repeated symbols is called a code word, i.e., a code word is what
is transmitted in place of one piece of information in the original
message. The set of all code words is called a code. If all the code
words in a code have the same length, then the code is called a
block code. The repeat codes are block codes.
 Detection and Correction

One feature that a useful code must have is the ability to detect
errors. The repeat code with code words having length 5 can
always detect from 1 to 4 errors made in the transmission of a code
word, since any 5 letter word composed of more than one letter is
not a code word. However, it is possible for 5 errors to go
undetected (how?). We would say that this code is 4-error
detecting. Another feature is the ability to correct errors, i.e., being
able to decode the correct information from the error riddled
received words. The repeat code we are dealing with can always
correct 1 or 2 errors, but may decode a word with 3 or more errors
incorrectly (how?), so it is a 2-error correcting code.
 Coding Needed

In the Mariner 9 mission, without coding and a failure probability

p = 0.05, 26% of the image would be in error ... unacceptably poor
quality for TV transmission.

Any coding will increase the length of the transmitted message.

Due to power constraints on board the probe and equipment
constraints at the receiving stations on Earth, the coded message
could not be much more than 5 times as long as the data.

Thus, a 6-tuple of data could be coded as a codeword of about 30

bits in length.
 Other concerns
A second concern involves the coding procedure. Storage of data
requires shielding of the storage media – this is dead weight aboard
the probe and economics require that there be little dead weight.
Coding should therefore be done “on the fly”, without permanent
memory requirements.

Finally, decoding needs to be done rapidly. The Jet Propulsion

Laboratory in Pasadena, California will process the signals and
reconvert them to picture images for the press which will be
gathered at JPL.

Besides this NASA priority, rapid decoding is needed so that

feedback to the probe becomes viable – redirecting the camera
based on what is seen.
 The Code
The 5-repeat code would satisfy the mission specs, but it is only
2-error correcting ... leaving 1% of the image in error.

The actual code selected is 7-error correcting and this reduced the
probability of error in the image to only 0.01%.

The decision on which code to use was based primarily on the

decoding algorithm. The algorithm was carried out by a fairly
simple piece of specialized circuitry called “The Green Machine.”

The code selected was the (32,6)-Biorthogonal Reed-Muller Code.

We will go through the construction and use of this code after
looking at some of the results.
 Results
"Inca City" is the informal name given by Mariner 9
scientists in 1972 to a set of intersecting, rectilinear
ridges that are located among the layered materials
of the south polar region of Mars. Their origin has
never been understood; most investigators thought
they might be sand dunes, either modern dunes or,
more likely, dunes that were buried, hardened, then
exhumed. Others considered them to be dikes
formed by injection of molten rock (magma) or soft
sediment into subsurface cracks that subsequently
hardened and then were exposed at the surface by
wind erosion.
Inca City:
-80 Lat., 64 Long.
 Inca City

The Mars Global Surveyor (MGS) Mars Orbiter

Camera (MOC) has provided new information
about the "Inca City" ridges, though the camera's
images still do not solve the mystery. The new
information comes in the form of a MOC red
wide angle context frame taken in mid-southern
spring. The MOC image shows that the "Inca
City" ridges, located at 82°S, 67°W, are part of a
larger circular structure that is about 86 km (53
mi) across.
 Inca City

It is possible that this pattern reflects an origin related to an ancient,

eroded meteor impact crater that was filled-in, buried, then partially
exhumed. In this case, the ridges might be the remains of filled-in
fractures in the bedrock into which the crater formed, or filled-in cracks
within the material that filled the crater. Or both explanations could be
wrong. While the new MOC image shows that "Inca City" has a larger
context as part of a circular form, it does not reveal the exact origin of
these striking and unusual martian landforms.
Inca City
 Hadamard Matrices

An n × n matrix H = hij is an Hadamard matrix of order n if the entries

of H are either +1 or -1 and such that HH = nI, where H is the transpose of
H and I is the order n identity matrix. Put another way, a (+1,-1)-matrix is
Hadamard if the inner product of two distinct rows is 0 and the inner product
of a row with itself is n.
A few examples of Hadamard matrices are;

1 1 -1 1 1 1 1 1 1 1
1 -1 1 -1 1 1 1 -1 1 -1
1 1 -1 1 1 1 -1 -1
1 1 1 -1 1 -1 -1 1
 Hadamard Matrices
These matrices were first considered as Hadamard determinants. They were
so named because the determinant of an Hadamard matrix satisfies equality in
Hadamard's determinant theorem, which states that if X = xij is a matrix of
order n where | xij | ≤ 1 for all i and j, then
| det X | ≤ nn/2

It is apparent that if the rows and columns of an Hadamard matrix are

permuted, the matrix remains Hadamard. It is also true that if any row or
column is multiplied by -1, the Hadamard property is retained.[Prove this]
Thus, it is always possible to arrange to have the first row and first column of
an Hadamard matrix contain only +1 entries. An Hadamard matrix in this
form is said to be normalized.
 Order of Hadamard Matrices
Theorem: The order of an Hadamard matrix is 1,2 or 4n, n an integer.

Proof: [1] is an Hadamard matrix of order 1 and the first example above is an
Hadamard matrix of order 2. Suppose now that H is an Hadamard matrix of
order h > 2. Normalize H and rearrange the first three rows to look like:
1 ..... 1 1 ..... 1 1 ..... 1 1 ..... 1
1 ..... 1 1 ..... 1 -1 .....-1 -1 .....-1
1 ..... 1 -1 .....-1 1 ..... 1 -1 .....-1
x y z w
Where x,y,z,w are the numbers of columns of each type. As the order is h,
x+y+z+w=h
and taking the inner products of rows 1 and 2, 1 and 3, and, 2 and 3 we get
x+y-z-w=0
x-y+z-w=0
x - y - z + w = 0.
Solving this system of equations gives,
x = y = z = w = h/4.
Thus, the integer h must be divisible by 4.
 Properties of Hadamard Matrices
Corollary: If H is a normalized Hadamard matrix of order 4n, then every row
(column) except the first has 2n minus ones and 2n plus ones, further n minus
ones in any row (column) overlap with n minus ones in each other row
(column).

Proof: This is a direct result of the above proof since any two rows other than
the first can take the place of the second and third rows in the proof. The same
argument can be applied to the columns.

Hadamard matrices are known for many of the possible orders, the
smallest order for which the existence of an Hadamard matrix is in doubt is
currently 668 (A solution for the previous unknown case of 428 was
announced by Kharaghani and Tayfeh-Rezaie in June 2004).
 Construction of Hadamard Matrices
While there are a great many construction methods for Hadamard
matrices, we will only consider one of the simplest, the direct product
construction.

Construction - Given Hadamard matrices H1 of order n and H2 of order m

the direct product of these two matrices, represented by:

h11H2 h12H2 ... h1nH2

h21H2 h22H2 ... h2nH2

... ... ... ...

hn1H2 hn2H2 ... hnnH2

where H = | hij| , is an Hadamard matrix of order nm.

 Example
Let H 1=H 2=
[1 1
1 −1
.
]
The construction gives:

[ ]
1 1 1 1
H 1∗H 2=
H 2
H 2 [ ]
H 2
−H 2
=
1
1
1
−1
1
−1
1
−1
−1
−1
−1
1
.
 Back to Mariner 9
Recall that in the Mariner 9 mission, the data consisted of binary 6-
tuples (64 grayness levels) and transmission restrictions permitted
coding that would lengthen the transmitted words to about 30 bits.

The 5-repeat code would satisfy this condition, but it is only 2

error correcting.

The code chosen was a Reed-Muller code. The code words are 32
bits long and there are 64 of them. The code words are the rows of
a 32 × 32 Hadamard matrix constructed by repeating our example
construction 3 more times, and their negatives.
 Error Correction
The ability of a code to correct errors is directly related to the
“distance” between code words. We must make this concept
precise.

The block codes we will talk about will be subsets of the set of
all n-tuples whose coordinates come from an alphabet of size k.
We will denote this large set by V(n,k) and will often think of it
as a “vector space”, but this is not accurate unless the alphabet is
a field. In this more general setting, we do not do algebraic
operations with our “vectors” since the alphabet need not have
any algebraic properties. When we are being careful, we will use
the term word instead of “vector”.
 Hamming Distance
The Hamming distance between two words in V(n,k) is the
number of places in which they differ.

So, for example, the words (0,0,1,1,1,0) and (1,0,1,1,0,0) would

have a Hamming distance of 2, since they differ only in the 1st and
5th positions. In V(4,4), the words (0,1,2,3) and (1,1,2,2) also have
distance 2.

This Hamming distance is a metric on V(n,k), i.e., if d(x,y)

denotes the Hamming distance between words x and y, then d
satisfies:
1) d(x,x) = 0
2) d(x,y) = d(y,x), and
3) d(x,y) + d(y,z) ≥ d(x,z). (triangle inequality)
 Hamming Distance

The first two of these properties are obvious, but the triangle
inequality requires a little argument (this is a homework problem).

Since we will only deal with the Hamming distance (there are other
metrics used in Coding Theory), we will generally omit the
Hamming modifier and talk about the distance between words.
 Minimum Distance
The minimum distance of a code C is the smallest distance
between any pair of distinct codewords in C. It is the minimum
distance of a code that measures a code's error correcting
capabilities. If the minimum distance of a code C is 2e + 1, then C
is a 2e-error detecting code since 2e or fewer errors in a codeword
will not get to another codeword and is an e-error correcting code,
since if e or fewer errors are made in a codeword, the resulting
word is closer to the original codeword than it is to any other
codeword and so can be correctly decoded (maximum-likelihood
decoding).

In the 5-repeat code of V(5,4) (codewords: 00000, 11111, 22222,

and 33333) the minimum distance is 5. The code detects 4 or fewer
errors and corrects 2 or fewer errors as we have seen.
 Our Reed-Muller Code
From the properties of a Hadamard matrix, we can see that two
different rows from the matrix will differ in exactly half of their
positions (since the dot product will be 0). If we take one of the
two rows and negate all its elements, the dot product will not
change, so again the two rows will differ in exactly half of their
elements. Finally, a row and its negation will differ in all
positions. Thus, the minimum distance between these rows is
half the length of the rows.

In our case, the code words are the rows of two 32 x 32

Hadamard matrices (one the negation of the other) so the
minimum distance is 16, which makes this a 7-error correcting
code.
 Encoding
Encoding:
As there are 64 code words and 64 data types, any assignment of
code word to data type will work, but the requirement that the
encoding should require no memory meant that an arbitrary
assignment would not do.

Using these special Hadamard matrices makes the code a 6

dimensional vector space, so there is a basis with 6 elements (any
linear combination of which gives a code word). The data type 6-
tuple is used to provide the coefficients for the linear combination
of the basis vectors ... thus associating a unique code word to each
data type.

This simple computation can be hard wired and requires no

memory.
 Decoding
Decoding:
As we have previously mentioned, the real reason for selecting
this code was that it had a very fast decoding algorithm which we
now describe.

First, convert all the code words (and the received vector) to ± 1
vectors by turning the 0's into -1's. Take the dot product of the
received vector with each of the code words in turn. As soon as the
result is 16 or greater, decode as that code word.

Suppose no errors have been made in transmission. Then the dot

product of the received vector with itself will be 32 and with any
other codeword will be 0 or -32.
 Decoding
Decoding:
This follows since the distance between two code words is the
weight of their difference (which is another code word) and so is
either 0, 16 or 32. If 0, the code words are the same. If 32, the code
words have no common component and the dot product of the ±1
form will be -32. In all remaining cases, 16 places are the same and
16 places are different, giving a dot product of 16 – 16 = 0.

For each error that occurs, the dot product will decrease by 2 (or
increase by 2 from an incorrect codeword). If no more than 7 errors
occur, the dot product with the correct code word decreases to at
least 18 and the dot product with incorrect code words increases to
at most 14 ... so correct decoding will occur. If 8 or more errors
occur, there will be dot products of at least 16 but correct decoding
is not possible.
 The Green Machine
Decoding:

Even though this is a rapid decoding algorithm, the

computations involved can be speeded up by a factor of 3
by using a Fast Fourier Transform for Abelian groups. This
is what was actually done by the “green machine”.
 Other Missions

The Voyager 1 & 2 spacecraft transmitted color pictures of Jupiter

and Saturn in 1979 and 1980. Color transmission requires 3 times
the amount of data, so a different code (the Golay (24,12,8) code)
was used. It is only 3-error correcting, but its transmission rate is
much higher. Voyager 2 went on to Uranus and Neptune and the
code was switched to a Reed-Solomon code for its higher error
correcting capabilities.