Cryptography Reference
In-Depth Information
P
F q
F n + + p
q
S
T
CA
F n +
q
F n + + p
q
Fig. 5. The Double-Layer Square Scheme
C◦ S, A◦ S ). See
figure 5 for a graphical representation. Note that all intermediate operations are
quadratic over
Now we can write the public key
P
of Square+ as
P
:= T
(
. If we leave out the embedding modifier for a moment
(transformation S ), there are two parts of Square+, namely the invertible, but
“soft” part X 2 , represented by transformation
F q ,asis
P
C
, and the not-invertible “hard”
part a 1 ,...,a p , represented by transformation
. If we manage to separate them,
we are done as there is an ecient attack against Square [2].
A
4.1 Odd-Characteristic HFE Attack against Square+
To this aim, we have a closer look at “odd Characteristic HFE” (or odd-HFE)
and its cryptanalysis [2, 6]. In particular we notice that the central map of odd-
HFE is
γ i,j X q i + q j
( i,j ) Δ ( D )
2 : i j, q i + q j
for a set of admissible degrees Δ ( D ):=
( i, j )
N
D }
N
the
set of non-negative integers and γ i,j F q n the coecients of the corresponding
private key. Setting D =2and γ (0 , 0) =1,weobtain Δ (2) = (0 , 0) and the central
map of odd-HFE coincides with the one of Square+. As a result, we can apply
the cryptanalysis of Bettale et al. [2] against odd-HFE also against Square+.
Alas, this cryptanalysis does not include the case odd-HFE+, so we need to
investigate this question closely to determine if we can break Square+ within
this framework. As we will see below, it works but there are subtle changes to
be made.
As for the original attack against odd-HFE, the key point is the observation
that we can write X 2 as a matrix of small rank over the extension field. More
to the point, we have X 2 = x F x over
{
,
for x =( X, X q ,X q 2 ,...,X q n 1 )
F q n +
with
F 1 , 1 = 1 but
F i,j
= 0 otherwise. As only
F 1 , 1 is non-zero, we obviously
have rank(
) = 1. A similar observation for a MinRank attack against HFE was
already used by Kipnis-Shamir [15]. Note that expressing X 2
F
over the ground
field yields a much higher rank, in practice close to ( n + ).
To ease notation and to mount the attack, we follow the approach of [2] and
start with the vector ( θ 1 ,...,θ n + )
n +
q n + . Note that this vector has a double
function: First, it fixes a basis of the vector space
F
n + q , i.e. over the ground
field, and second, the elements θ 1 ,...,θ n + are simultaneously interpreted over
F
Search WWH ::




Custom Search