2013 Argentine School of Micro-Nanoelectronics, Technology and Applications

Pseudorandom Number Generator Based on the

Residue Number System and its FPGA
Implementation
Carlos Arturo Gayoso, C. González, L. Arnone, M. Rabini, Jorge Castiñeira Moreira


II. RESIDUE NUMBER SYSTEM
Abstract— In this paper a novel Pseudorandom Number
Generator is presented. It is based on the Residue Number Arithmetic circuits based on complement-2 operations like
System (RNS), which allows us to design a very fast circuit that adders for instance, have to propagate carry bit information
has a very different way of operating with respect to other from the less significant bit to the most significant bit, which
generators. A set of different tests is used to determine the quality is related to a degradation in performance as the number of
of the performance of the proposed generator. Results are shown carries rises. On the other hand, RNS [1],[2],[3],[4] is an
for an FPGA implementation of the proposed generator. efficient technique to overcome this problem, since operates
over independent channels that do not exchange information.
Index Terms—pseudorandom number generator, residue
Thus RNS based schemes are constituted of several channels,
number system, programmable logic.
each one operating using a reduced number of bits.
Arithmetic circuits of t bits can be transformed into
channels of bits each [5], by using
I. INTRODUCTION RNS (See Fig. 1). This characteristic makes RNS very suitable

I N many practical applications such as cryptography, model

simulation, sampling, games of chance, among others, there
is a need of the generation of series of random numbers. This
for implementing, for instance, digital signal processing (DSP)
applications [6],[7],[8].

is achieved, for example, by means of tables, specific

algorithms or electronic circuits. Unlike the natural sources of
noise, these generators possess a finite period, so that they are
called pseudorandom number generators (PRNGs).
There are several metrics for determining the quality, in
terms of randomness, of the generated sequence. These
metrics offer a quantitative measure of how similar to a real
random sequence is the generated pseudorandom sequence. In
this paper a new electronic circuit of a pseudorandom
generator is proposed. A set of classic tests, the Diehard test, Fig. 1. General scheme of an RNS circuit.
the statistic complexity test and the Hurst exponent test are
used to provide a measure of the quality of the randomness of A given scheme based on RNS is defined by a set of
the proposed pseudorandom number generator. relative prime numbers m = {m1, m2..., ml}, called modules.
The number of different quantities that can be represented is
M = mi (i = 1, 2, …, l), which is it dynamic range. Thus, any
integer number 0  X < M is represented by a set of l residues
[x1, x2, ..., xl], where xi = X mod mi (i = 1, 2, …, l). The best
This paper was submitted for review on April 24, 2012 and was supported advantage of using RNS is its capability of performing sums,
by the Universidad Nacional de Mar del Plata, and took place in the
Laboratory of Electronic Components and Communications Laboratory.
subtractions and multiplications at very high speed. This is
C. A. Gayoso is a member of the of the Laboratory of Electronic because RNS is defined over a ring of integers modulo M such
Components at the National University of Mar del Plata (corresponding that:
author; e-mail: [email protected]).
C. M. González is a member of the of the Laboratory of Electronic
Components at the National University of Mar del Plata.
(1)
L. J. Arnone is a member of the of the Laboratory of Electronic
Components at the National University of Mar del Plata. Where the symbol * denotes sum, subtraction or
M. Rabini is a member of the of the Laboratory of Electronic Components multiplication modulo M. (1) makes clear the main advantage
at the National University of Mar del Plata.
J. C. Moreira is a member of the of the Communications Laboratory of at
of RNS, since modulo M operations are done over l
the National University of Mar del Plata (e-mail: [email protected]). independent channels. The total throughput of the scheme is

ISBN 978-987-1907-44-1 9 IEEE Catalog Number: CFP1354E-CDR

Authorized licensed use limited to: Franco Fabian Tolosa. Downloaded on September 15,2023 at 18:13:18 UTC from IEEE Xplore. Restrictions apply.
 2013 Argentine School of Micro-Nanoelectronics, Technology and Applications

given by the throughput of each channel in modulo mi, since position of the binary representation will not be equally likely.
there is no shared information among channels. In spite of Let us see the following example in TABLE I, where if M =
operations like division and comparison can be difficult to 11, then t = 4.
implement, this does not limit the use of RNS, and in fact
digital signal processing (DSP) became one of the main
application fields for RNS. Thus, very common DSP
operations like sums and multiplications can be performed in a
faster way if they are implemented using RNS, as being shown
for applications like discrete transforms, digital filtering or
image processing [9],[10],[11].

III. PROPOSED RNS BASED INVERSE CONGRUENTIAL

GENERATOR (RNS-ICG)
Equation (2) shows the so-called Inverse Congruential
Generator (ICG). The value x-1 is the multiplicative inverse of
x modulo M, that is (x–1 . x )M = 1, for x  0 and x–1 = 0 if x =
0; where a and b are constants.

(2)

The proposed circuit is based on this algorithm and is

called Residue Number System-Inverse Congruential
Generator (RNS-ICG). The idea is not to solve (2) in the RNS
but the introduction of this system will give, in addition to
speed, a different dynamic. The proposed RNS-ICG is
presented in two different forms, called type I and II. Fig. 2
shows the corresponding circuit for l channels, each one Fig. 2. General scheme of the proposed generator and retail of one of the
channels.
operating with hj bits. Each channel is a small ICG that sends
and receives information from the other channels.
Expression (3) is used instead of (2) for obtaining better As is seen in TABLE I, and as an example, bit in position
statistics properties. In (3) gj and mj are the primitive root and b2 has 9 zeroes and 4 ones. Thus, values ‘0’ and ‘1’ are not
the modulo of each channel respectively, and r j, i is the residue equally likely in each position, and this is not a desirable
of the channel j for iteration i. characteristic of a pseudorandom number generator. Even
when decimal representation of the values can be equally
rj , i  g j rj , i  2  b j , i  mod m j
likely, this does not imply a uniform Probability Density
(3) Function (PDF) of the binary values. Three different
approaches called A, B and C, are adopted here for
In order to obtain an increased level of randomness, overcoming this problem.
channels are allowed to interact, such that parameter b in (2) is
replaced by bj,i in (3) and is no longer constant. For generator TABLE I
EXAMPLE FOR M = 11 AND t = 4
type I, referred here as RNS-ICG-I, bj,i for channel j and
b3 b2 b1 b0 b3 b2 b1 b0
iteration i is calculated as:
0 0 0 0 0 7 0 1 1 1
k  l 1 1 0 0 0 1 8 1 0 0 0
bj, i  r
k 0
k , i 1
(4) 2
3
0
0
0
0
1
1
0
1
7
8
0
0
0
0
0
0
0
1
k j 4 0 1 0 0 9 0 0 1 0
For generator type II, referred here as RNS-ICG-II, if  5 0 1 0 1 10 0 0 1 1
6 0 1 1 0
denotes bitwise Exclusive-OR operation, we have:

k l 1 A. Approach A
b j ,i  r
k 0
k , i 1
(5)
t+1
In this approach the set m is selected such that 2t < M <
k j 2 . For the case t = 32 for instance, set m can be equal to m =
{3, 11, 17, 19, 23, 29, 31, 37} which makes M =
Since in the RNS with l modules numbers of t bits in the 8,154,657,291 > 232 = 4,294,967,296. This means that the
range 2t < M < 2t+1 are obtained, M is not an exact power of 2. desired range can be properly represented. Only those values
Therefore, in order to operate with t bits, a set of l modules is that are less than 232 are taken, the rest are discarded. If the
selected to be always larger than 2t, which in turn is related to generated sequences have a uniform PDF in the range 0 to M –
the following inconvenient. Even when numbers read in 1, the corresponding binary sequences have a uniform PDF in
decimal form are equally likely, values ‘0’ and ‘1’ in each the range 0 to 232 – 1. As a consequence of the discarding

ISBN 978-987-1907-44-1 10 IEEE Catalog Number: CFP1354E-CDR

Authorized licensed use limited to: Franco Fabian Tolosa. Downloaded on September 15,2023 at 18:13:18 UTC from IEEE Xplore. Restrictions apply.
 2013 Argentine School of Micro-Nanoelectronics, Technology and Applications

procedure, there will not be a generated number at each new value, taken from the bus d, is stored. This means that its
iteration. value is statistically modified in average 1 time every 32
iterations, for the case of using the 5 less significant bits of
B. Approach B Register C. When the bit in position 32 that is output by the
In this approach the set m is selected such that the RNS to binary converter is equal to one, bits 31 to 0 are
subtraction M - 232 is minimal. This means that the dynamic replaced by the OR-Exclusive operation between these bits
range is of 33 bits, and we simply discard the most significant and the value stored at Register E. This value stored in
bit. The throughput is enhanced because we generate a number Register E had been stored in a previous moment. The greater
at each iteration. The advantage of this approach with respect is the number of bits selected to compare in Register C, the
to approach A is that approach B is faster, does not need a statistically further away from the actual value is the stored
comparator circuit, and each iteration generates a number. In value. Bitwise operation Exclusive-OR is applied due to its
this paper, 32-bit generators were tested, with modules up to 7 implementation simplicity.
bits ({3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43, 47, 53, 59,
61, 67, 71, 73, 79, 83, 89, 97, 101, 103, 107, 109, 113, 127})
and taken in groups of 5, 6, 7 and 8. TABLE II shows the best
results from combinations of modules so that the difference M
- 232 in less than 0.01 %.
If the first set m = {3, 43, 47, 67, 97, 109} is selected, as done
in this paper, the uniformity of the corresponding PDF is
altered. There will be some numbers more likely than others.
Thus for instance, we can obtain 110 from
000000000000000000000000000000001 2 or from
1000000000000000000000000000000012 but these cases are
only a 0.000171% of the total. In practice values ‘0’ and ‘1’
are then equally likely at each position and the corresponding
PDF is very close to be uniform, and this is more accurate as
the number of bits of the representation used is greater.
TABLE II
SET OF MODULES THAT ARE CLOSER TO 232 IN EXCESS Fig. 3. Correction circuit.

M – 232 % m0 m1 m2 m3 m4 m5 m6
IV. RANDOMNESS TESTS
0.000171 % 3 43 47 67 97 109
0.000329 % 5 11 13 17 53 59 113
Sequences generated by the proposed PRNG are compared
to other known and tested PRNGs [12], [13]. They are:
0.001602 % 3 7 13 37 47 83 109
0.001765 % 5 11 17 23 29 71 97  Uniform, generated by MatLab 7.01.
0.001930 % 11 23 53 59 61 89  MWCG, Multiply With Carry Generator, with the
0.002493 % 5 11 13 19 61 71 73 program Makewhat of the pack Diehard.
0.002979 % 3 11 13 23 67 73 89  MTHR4, “the mother of all random number
0.003663 % 7 31 41 43 103 109 generators” of the pack Diehard.
0.003954 % 3 7 13 37 53 71 113
 SWBMWC, a combination of generators Subtract With
Borrow and Multiply With Carry, of the pack Diehard.
0.004044 % 13 17 37 61 79 109
0.004114 % 13 19 53 59 67 83 We have used 32 bits generators. With m = {3, 11, 17, 19,
0.005759 % 3 7 13 19 71 107 109 23, 29, 31, 37} for approach A, and m = {3, 43, 47, 67, 97,
109} for approaches B and C. Coefficients operating in each
C. Approach C kind of generator are the primitive roots g = {2, 2, 3, 2, 5, 2, 3,
In this case restrictions applied in approaches A and B are 2} in the first case, and g = {2, 3, 5, 2, 5, 6} for the second
eliminated, a number is generated at each iteration and the and third approach.
corresponding PDF is uniform. As done in approach B, a set m Basic tests like autocorrelation, spectral analysis and
is selected such that the difference M - 232 is less than 0.01 % uniformity tests were applied to the proposed PRNG, and the
if numbers of 32 bits are desired, so we need to work with 33 well known PRNGs mentioned above. There were not
bits. As seen in Fig. 3, following the pseudorandom number significant differences in performance of the proposed PRNG
generator (PRNG) a circuit, called the correction circuit, is with respect to the known PRNGs taken as reference.
connected. If the bit in position 32 that is output by the RNS to Results presented in this work were obtained by simulation
binary converter is equal to zero, bits 31 to 0 are output at bus of the circuits. Nevertheless these conclusions have been also
d. Information stored at Register E is periodically updated. If a validated by using the tool SignalTap of Altera. SignalTap
given number of for instance the less significant 5 bits of Embedded Logic Analyzer (ELA) is a logical analyzer that
Register C are equal to an arbitrary predetermined value, a 1
Copyright 1984-2004, The MathWorks Inc.

ISBN 978-987-1907-44-1 11 IEEE Catalog Number: CFP1354E-CDR

Authorized licensed use limited to: Franco Fabian Tolosa. Downloaded on September 15,2023 at 18:13:18 UTC from IEEE Xplore. Restrictions apply.
 2013 Argentine School of Micro-Nanoelectronics, Technology and Applications

can capture signals in real time, store them and then show As seen in TABLES III, IV and V the proposed PRGNs
them for their study. satisfactorily pass one of the most exigent randomness
analyses to determine the randomness of a given numerical
series.
V. TEST DIEHARD
The pack Diehard [14], [15] is a set of 17 tests developed VI. COMPLEXITY TEST
by George Marsaglia since 1993, and published in 1995. The A given system is said to be complex if there is no structure
Diehard software has as input a file of at least 11 Mbytes or behavior that can allow us to model it [14],[15],[16],[17].
containing the sequence of numbers to be analyzed in terms of From this point of view, complexity is used to measure the
randomness. 229 values, called p values, are obtained, each randomness of a given series. This concept easily understood is
one has to be in the range 0.025 < p < 0.975 to considering the however difficult to quantify. Due to this, several metrics were
sequence as random. Finally, the analysis determine the p proposed. Two metrics were utilized: a) statistical complexity,
value of these 229 values, that also must be in the range 0.025 and b) compression complexity.
< p < 0.975.
10 series of approximately 3,000,000 points of 32 bits A. Statistical Complexity
were generated for each PRNG. Results are shown in TABLE The statistical complexity metrics used in this paper are the
III, TABLE IV and TABLE V. López-Mancini-Calvet (CLMC) metric [18] and the Martin-
Plastino-Rosso (CMPR) metric [19].
TABLE III The CLMC metric is based on concepts like order,
p VALUES CALCULATED USING DIEHARD FOR 10 SERIES AND FOR information and equilibrium and defines statistical complexity
GENERATORS RNS-ICG-I Y RNS-ICG-II TYPE A of a system as the distance to the equally likely distribution of
Simulation
RNS-ICG-I RNS-ICG-II all states of the system. It is calculated as:
p p
0 0.759408 0.938063
 N
1
2
0.070830
0.363415
0.094446
0.353138   pi ln  pi    2

  p  1 
N
   

i 1
3 0.732662 0.105207 C LMC  i  
4 0.834524 0.399985 ln N  i 1  N 
5 0.102359 0.622250
 
6 0.836744 0.134909
  (6)
7 0.799920 0.201867
8 0.364214 0.659446
9 0.194649 0.569665
with pi the probability of each state and N the number of
states.
Expression (6) is the same as to evaluate an Euclidean
TABLE IV distance, but it does not take into account the random nature of
p VALUES CALCULATED USING DIEHARD FOR 10 SERIES AND FOR the PDF. In [19] is defined CMPR as:
GENERATORS RNS-ICG-I Y RNS-ICG-II, TYPE B
RNS-ICG-I RNS-ICG-II  N 
Simulation
p p   pi ln  pi  
  1  N
1 
0 0.653108 0.209772 CMPR  H DMPR  
i 1

ln N   1 
arccos   pi
N


1 0.534267 0.165843   arccos    i 1 
2 0.283026 0.516257    N 

3 0.488090 0.106113
4 0.755365 0.319848
(7)
5 0.267669 0.528632
6 0.619149 0.497950 In expression (7) DMPR involves the statistical distance
7 0.412185 0.229261 rather than the Euclidean distance. Statistical distance between
8 0.773033 0.091080 two points in a probability space of dimension N is the shortest
9 0.562224 0.050897
statistical length that connects these two points in that space.
The statistical length of a curve is the maximum number of
TABLE V
distinguishable points along that curve. Its mathematical
p VALUES CALCULATED USING DIEHARD FOR 10 SERIES AND FOR
GENERATORS RNS-ICG-I Y RNS-ICG-II TYPE C
expression is given by:
RNS-ICG-I RNS-ICG-II
Simulation
p p 1  N 1 
0 0.469149 0.010513 DMPR  arccos   pi 

 1   i 1 N 
1 0.149563 0.428405 arccos  

2 0.639931 0.149411
 N
3 0.123876 0.225993 (8)
4 0.818532 0.602838
5 0.151925 0.024404
As seen in TABLES VI and VII values CLMC and CMPR
6 0.050745 0.396904
7 0.133014 0.626420 obtained for the proposed PRNGs are very close to those of
8 0.049635 0.618624 the reference PRNGs.
9 0.942691 0.680900

ISBN 978-987-1907-44-1 12 IEEE Catalog Number: CFP1354E-CDR

Authorized licensed use limited to: Franco Fabian Tolosa. Downloaded on September 15,2023 at 18:13:18 UTC from IEEE Xplore. Restrictions apply.
 2013 Argentine School of Micro-Nanoelectronics, Technology and Applications

TABLE VI compression is 0%, that is, the algorithm can not detect any
LÓPEZ-MANCINI-CALBET COMPLEXITY MEASURED FOR TWO kind of regularity.
HISTOGRAMS, ONE OF 65,536 DISTINCT VALUES OTHER OF 1,000
TABLE VIII
CLMC
Generator
Czipping FOR THE PROPOSED RNS-ICG
Histogram Histogram
0 to 65535 0 to 1000 Czipping
RNS-ICG-I 1.6702x10-7 2.4566x10-7 Class A 1.0001622
A
RNS-ICG-II 1.6518x10-7 2.2599x10-7 Class B 1.0001621
RNS-ICG-I 1.6764x10-7 2.2696x10-7 Class C 1.0001622
B
RNS-ICG-II 1.6722x10-7 2.3429x10-7
RNS-ICG-I 1.6757x10-7 2.3442x10-7
C
RNS-ICG-II 1.6675x10-7 2.2150x10-7 VII. HURST EXPONENT
Uniform 1.6629x10-7 1.6887x10-7
MWCG 1.7963x10-7 2.4893x10-7 H. E. Hurst [22],[23] was employed at the project of the
MTHR4 1.8052x10-7 2.3899x10-7 dam of the Nilo River at the beginning of the 20th century.
SWBMWC 1.8018x10-7 2.2474x10-7 He developed a statistical analysis for measuring the volume
of a water reservoir in order to avoid both its empting and its
TABLE VII overflowing in spite of the fluctuations of the flow of the river.
MARTIN-PLASTINO-ROSSO COMPLEXITY MEASURED FOR TWO He based his analysis on the Einstein equation R = T0.5,
HISTOGRAMS, ONE OF 65,536 DISTINCT VALUES OTHER OF 1,000 developed for the analysis of the Brownian movement of
CMPR particles, where R is the distance travelled by the particle, and
Generator Histogram Histogram T the time between observations. Hurst discovered that this
0 to 65535 0 to 1000 expression, applied only to zero mean value and unitary
RNS-ICG-I 0.0335 0.0052 variance series can be generalized as follows:
A
RNS-ICG-II 0.0333 0.0050
RNS-ICG-I 0.0335 0.0050
B
RNS-ICG-II 0.0335 0.0051 R 
C
RNS-ICG-I 0.0335 0.0050    c . n H (11)
RNS-ICG-II
Uniform
0.0334
0.0334
0.0047
0.0042
S n
MWCG 0.0343 0.0052
MTHR4 0.0344 0.0051 Where R is the rank, S the standard deviation, c is a constant, n
SWBMWC 0.0344 0.0049 the number of points of the series and H the Hurst exponent,
with values between 0 and 1. When H = 0.50 the process is
independent, or random, independently of the type of
B. Compression Complexity
distribution. When the Hurst exponent is in the range 0.50 < H
Compression algorithms usually used for compressing files
≤ 1 the series is said to be persistent and it presents long term
are proposed in [20] as a tool for measuring the complexity of a
memory effects, so that what happens in the present time
given system. The Shannon-McMillan theorem states that for a
affects in the future. If at a given time the series shows for
given sequence of values there exists a maximum level of
instance consecutive ascendant values, it is very likely that the
compression without losing information. This limit is
next value will also be ascendant, and vice versa. Finally, when
proportional to the entropy of the sequence. A good measure of
the Hurst exponent is in the range 0 ≤ H < 0.5 the series is said
the randomness of a sequence is to see how much can be
to be antipersistent, with memory effects, and with more
compressed.
frequent slope changes than in the case of the Brownian
Classic compression programs of files can be used for this
movement.
analysis in order to quantify their entropy. A given
compression algorithm is optimum if it can achieve the level of TABLE IX
compression stated by the Shannon-McMillan theorem. One of HURST EXPONENT FOR THE PROPOSED AND REFERENCE PRNGs
the most utilized algorithms in file compressors is the Lempel-
Ziv algorithm [21], which is able to asymptotically reach the Generator H
Shannon-McMillan limit. WinZip2 9.0 uses an improved RNS-ICG-I 0.51028
A
version of the Lempel-Ziv algorithm called the Lempel-Ziv- RNS-ICG-II 0.52445
Welch algorithm. Compression complexity is defined as: RNS-ICG-I 0.50739
B
RNS-ICG-II 0.52119
RNS-ICG-I 0.50670
Nº of bits compressed file (10)
C
RNS-ICG-II 0.52476
Czippig 
Nº of bits uncompressed file Uniform 0.52689
MWCG 0.52616
MTHR4 0.52328
Results are shown in TABLE VIII. It can be seen that in all SWBMWC 0.52253
cases Czipping is slightly greater than 1. This is because of the
information added by the compression program. Obtained A desired value for PRGNs is a Hurst exponent close to H
 0.50 in order to determine that there is no interdependence
2
Copyright 1991-2004 WinZip Computing Inc.

ISBN 978-987-1907-44-1 13 IEEE Catalog Number: CFP1354E-CDR

Authorized licensed use limited to: Franco Fabian Tolosa. Downloaded on September 15,2023 at 18:13:18 UTC from IEEE Xplore. Restrictions apply.
 2013 Argentine School of Micro-Nanoelectronics, Technology and Applications

among sequence values. TABLE IX shows good values of this [10] J. Ramírez, A. García, P. Fernández, L. Parrilla, and A. Lloris,
“RNSFPL Merged Architectures for Orthogonal DWT”, Electronics
parameter for the studied PRGNs. Letters, vol. 36, no. 14, pp. 1198–1199, 2000.
[11] W. Chren, “RNS-Based Enhancements for Direct Digital Frequency
VIII. FPGA IMPLEMENTATION Synthesis”, IEEE Transactions on Circuits and Systems II, vol. 42, no.
8, pp. 516–524, 1995.
Since adders and multipliers circuits are the main
[12] M. Alioto, S. Bernardi, A. Fort , S. Rocchi and V. Vignoli, “On the
components of the hardware implementation of the RNS, an suitability of digital maps for integrated pseudo-RNGs”, Proc. ECCTD
exhaustive analysis was previously done about this sort of Cracow, Poland, Sep. 2003, p. III/349.
operations in the RNS, which were presented in [24] and [25]. [13] J. Savir “A new empirical test for the quality of random integer
In these papers different adders and multipliers previously generators”, IEEE Trans. Comput., vol. C-32, pp. 960, Oct. 1983.
presented in several papers like [26],[27],[28],[29],[30] were [14] C. M. González, H. A. Larrondo, C. A. Gayoso, L. J. Arnone,
analyzed. A circuit for the RNS-ICG-I class A was developed. “Implementación de Sistemas Caóticos en Dispositivos Lógicos
Programables”, XI Workshop de Iberchip. http://www.iberchip.org
The best adders and multipliers studied in [24] and [25] were /iws2005/index.html. 28 a 30 de Marzo de 2005. Salvador de Bahía,
selected for each case. The FPGA implementation uses a Brasil, pp. 339-340.
FLEX10K20RC240-4 [31] from Altera Corporation. The [15] H. Larrondo, C. González, M. Martín, A. Plastino, O. Rosso, “Intensive
operation frequency is 30.80 MHz, it requires 1073 logic cells Statistical Complexity Measure of pseudorandom bit generators”,
Physics A 356 (2005) 133–138.
(LC) and 32 memory blocks. On the other hand the circuit that
[16] A. Rosso, H. Larrondo, M. Martín, A Plastino, “Generalizad statistical
performs operation described in (2), with the same M, but complexity measure: a new tool for dynamical systems”, Net-Works
utilizing conventional binary arithmetic operates at 1.05 MHz, 2008, 9 al 11 de junio de 2008, Pamplona, España.
and requires 1902 LC. These two circuits operate in a very [17] C. González, H. Larrondo, C. A. Gayoso y L. Arnone, “Generación de
different way. secuencias pseudo aleatorias binarias por medio de un mapa caótico
3D”, IX Workshop Iberchip. http://www.iberchip.org
/IX/pages/index.html, 26 al 28 de marzo de 2003. Habana, Cuba. pp. 44
IX. CONCLUSIONS [18] López-Ruiz, R, Mancini, H. L, Calvet, X, “Astatistical measure of
complexity”, Physics Letters A 209, pp. 321-326, Dic. 1995.
In this paper a new PRNG based on the RNS is presented. [19] M. Martin, A. Plastino, O. Rosso, “Statistical complexity and
In this paper we present a new RNS-based PRNG. The disequilibrium ”, Physics Letters A 2003.
proposed circuit has a dynamic, due to the use of RNS, [20] A. Baronchelli, E. Caglioti, V. Loreto, “Measuring complexity with
zippers”, European Journal of Physics, vol 26 pp S69 – S77, 2005.
different than known generators. It also has a superior
performance in terms of speed and resource usage with respect [21] J. Fernández, H. Larrondo, A. Figliola, E. Serrano, J. Rostas, M. Hunter,
O. Rosso, “Brain maturation changes characterized by Algorithmic
to its realization in 2's complement. Rigorous tests like those Complexity (Lempel and Ziv Complexity)”, American Institute of
of the pack Diehard for testing randomness of numeric series Physics Conference Proceedings 913, ISBN 978-0-7354-0421-2, pp
have been applied to the generated sequences, to verify that 196-202, 2007.
the proposed generator satisfactorily passes the different tests [22] E. Peters, Chaos and order in the capital markets. A new view of cycles,
prices and market volatility, John Wiley and Sons Inc. 1991.
applied.
[23] E. Peters, Fractal market analysis. Aplying chaos theory to investment
Work continues on the implementation of other PRNGs and economics, John Wiley and Sons Inc. 1995.
based on the technique presented. [24] C. A. Gayoso, A. García, C. M. González, L. Arnone, J. C. García, E.
Boemo, “Estudio sobre el diseño de sumadores en aritmética de residuos
en lógica programable”, II Jornadas sobre Computación Reconfigurable
y Aplicaciones. 10 al 20 de septiembre de 2002, Almuñecar, Granada,
REFERENCES España. Anales pág 203. ISBN: 84-600-9928-8.
[1] N. Szabo and R. Tanaka, Residue Arithmetic and Its Applications to [25] C. Gayoso; C. González; M. Rabini; L. Arnone. “Study of Multipliers in
Computer Technology. McGraw-Hill, NY, 1967. Residue Arithmetic using Programmable Logic”, IEEE Catalog Number:
[2] M. Soderstrand, W. Jenkins, G. Jullien, and F. Taylor, Residue Number CFP1254E-CDR, ISBN: 978-987-1620-93-7, pp 17-21, EAMTA 2012.
System Arithmetic: Modern Applications in Digital Signal Processing. [26] M. A. Bayoumi and G. A. Jullien, “A VLSI Implementation of Residue
IEEE Press, 1986. Adders”, IEEE Transactions on Circuits and Systems, vol. 34, no. 3, pp.
[3] F. Taylor, “Large moduli multipliers for signal processing”, IEEE 284-288, Mar. 1987.
Transactions on circuits and systems, vol. 28, no. 7, pp. 731–736, 1981. [27] M Dugdale, “VLSI Implementation of Residue Adders based on Binary
[4] F. Taylor, “Residue arithmetic: A tutorial with examples”, Computer Adders”, IEEE Transactions on Circuits and Systems II: Analog and
Magazine, IEEE, vol. 1, pp. 59–62, 1984. Digital Signal Processing, vol. 39, no. 5, pp. 325-329, Mar. 1987.
[5] J. Savir, “A new empirical test for the quality of random integer [28] G. A. Jullien, “Implementation of Multiplication, Modulo a Prime
generators”, IEEE Trans. Comput., vol. C-32, pp. 960, Oct. 1983. Number, with Applications to Number Theoretic Transforms”, IEEE
[6] G. A. Jullien, “Implementation of Multiplication, Modulo a Prime Transactions on Computers, vol. C-29, no. 10, pp. 899-905, Oct. 1980.
Number, with Applications to Number Theoretic Transforms”, IEEE [29] J. C. Smith and F. J. Taylor, “A fault-tolerant GEQRNS processing
Transactions on Computers, vol. C-29, no. 10, pp. 899-905, Oct. 1980. element for linear systolic array DSP application”, IEEE Transactions
[7] J. C. Smith and F. J. Taylor, “A fault-tolerant GEQRNS processing on Computers, vol. 44, no. 9, pp. 1121-1130, Sep. 1995.
element for linear systolic array DSP application”, IEEE Transactions [30] J. Ramírez, P. G. Fernández, U. Meyer-Bäse, F. J. Taylor, A. García and
on Computers, vol. 44, no. 9, pp. 1121-1130, Sep. 1995. A. Lloris, “Design of Index-based RNS-DWT Architectures for Custom
[8] J. Ramírez, P. G. Fernández, U. Meyer-Bäse, F. J. Taylor, A. García and IC Designs”, Proc. of 2001 IEEE Workshop on Signal Processing
A. Lloris, “Design of Index-based RNS-DWT Architectures for Custom Systems SiPS’2001, pp. 70-79, Sep. 2001.
IC Designs”, Proc. of 2001 IEEE Workshop on Signal Processing [31] Altera Corporation, http://www.altera.com/literature/ds/dsf10k.pdf,
Systems SiPS’2001, pp. 70-79, Sep. 2001. FLEX10K Embedded Programmable Logic Device Family Data Sheet,
[9] W. Chren, “RNS-Based Enhancements for Direct Digital Frequency 2001.
Synthesis”, IEEE Transactions on Circuits and Systems II, vol. 42, no.
8, pp. 516–524, 1995.

ISBN 978-987-1907-44-1 14 IEEE Catalog Number: CFP1354E-CDR

Authorized licensed use limited to: Franco Fabian Tolosa. Downloaded on September 15,2023 at 18:13:18 UTC from IEEE Xplore. Restrictions apply.