Available online at www.sciencedirect.

com
Available online at www.sciencedirect.com

Energy
Procedia

Energy Procedia
(2011)16
000000
Energy 00
Procedia
(2012) 645 650
www.elsevier.com/locate/procedia

2012 International Conference on Future Energy, Environment, and Materials

A New Method to Sampling of Multiple Bandpass Signals

Luo Junyi*; Tian Shulin; Wang Zhigang; Guo Lianping
School of Automation Engineering of University of Electronic Science and Technology of China, Chengdu 611731,
China

Abstract
A traditional assumption underlying most data converters is that the signal should be sampled at a rate exceeding twice the highest
frequency. In this paper, we employ a method for low-rate sampling of multi- band signals via applying periodic nonuniform
sampling in shift-invariant spaces generated by m kernels with period T. To avoid the non-ideal filter in traditional method, we use
reconstruction functions to recovery the sampled signal completely. Finally, we validate the method in MATLAB; the conclusion of
simulation shows the frame-work presented here is feasible.

2011 Published by Elsevier B.V. Selection and/or peer-review under responsibility of International Materials Science Society.

2011 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of FEEM2012

Open access under CC BY-NC-ND license.

Keywords: periodic nonuniform sampling; shift-invariant spaces; multi- band signals; reconstruction functions

1 Introduction
One goal in designing a software defined radio (SDR) receiver is to move the analog-to-digital
converter (ADC) as close as possible to the antenna. To achieve this goal, one can manage to use a
wideband high-speed ADC to convert the RF signals to digital signals. With the development of wireless
technology, this enables the modulation of narrow-band signals by high carrier frequencies. To
demodulate the desired signals, the required sampling rate for the ADC could often be too high to be
attained if the Nyquist sampling theorem is to be satisfied [1]. The uniform bandpass sampling method
has been proposed to figure out the problem [2] , and this is a promising way for multi-band radio
communication. The uniform Bandpass sampling is the intentional aliasing of the information bandwidth
of the signal [3]. The sampling frequency requirement is no longer based on the frequency of the RF
carrier, but rather on the information bandwidth of the signal. Thus, the resulting processing rate can be
significantly reduced. However, the uniform sampling still suffers from many constraints such problem of
*

Corresponding author. Tel.: 02861831307

E-mail address: luojunyi2009@qq.com

1876-6102 2011 Published by Elsevier B.V. Selection and/or peer-review under responsibility of International Materials Science Society .
Open access under CC BY-NC-ND license. doi:10.1016/j.egypro.2012.01.104

646

Luo name
Junyi et
al. / Energy
Procedia
16 (2012)
645 650
Author
/ Energy
Procedia
00 (2011)
000000

timing jitter in A/D conversion process [4]. For nonuniform samples, there are both iterative methods and
noniterative methods to recreate the signals; these methods presuppose exact knowledge of the sample
locations. This is not always the case, and there may occur situations where the location data is
unavailable or partially available.
In [5], a sample in shift-invariant spaces was proposed to overcome these problems. The reconstruction
of sampled signals is achieved by forming linear combinations of a set of reconstruction function that
span a subspace; such functions can be expressed as linear combinations of shifts of a set of generators
with period T. This model encompasses many signals used in communication and signal processing. Any
signal x(t) in a SI space generated by m functions shifted with period T can be perfectly recovered from m
sampling sequences, obtained by filtering x(t) with a bank of m filters and uniformly sampling their
outputs at times nT. However, this method needs to use a digital correction filter, which is difficult in
achievement in digital electrocircuit. Our research is focus on the samples that are from known shiftinvariant spaces to avoid the non-ideal filter in traditional method.
This paper is organized as followed. Section II sets up the sampling model. In Section III, we discuss
the reconstruction of sampled signals. In finally, section IV shows simulation results.
2 Proposed Scheme
The architecture of nonuniform sampling system is shown in figure 1.

Fig 1 the mode of periodic non-uniform sampling

The nonuniform sampling process converts a continuous analogue signal x(t)L2-space into its
discrete representation, the architecture of periodic nonuniform sampling system is shown in figure 1.
Let ai(t) as one of s nonuniform sample sequences,

=
ai (t )

(t Tn i )(0 i s 1)

(1)

n =

Where, T is the sampling period, is sequence separation.

One of s sampled functions,
n=

yi (t )= x ( nT + i ) (t nT i )

(2)

n =

Where, 0 i s 1 .and the corresponding spectra is given by

1 +
=
Yi ( )
X ( 2 n / T )e j 2 ni / T
T n =

m1

z r [n] (t nT ) : r [n] L }

=
V ( ) {

=
p 0 n

In order to reconstruct x(t) from these samples y[n]( y[n]=[ y0[n], y1[n],,ys-1[n]]), it is assumed that

(3)

647

LuoAuthor
Junyi name
et al. // Energy
650
Energy Procedia
Procedia 16
00 (2012)
(2011) 645
000000

x(t) lies in a subspace V()of L2 . In this paper, we define that the V() are generated by m space
functions (t)
We can represent any x(t)V() as follow
x (t ) =

m 1

r [n]

=
p 0 nz

p (t

(4)

nT )

The only restriction on the choice of the function train{p(t)} is for guaranteeing a unique stable
representation of any signal in V() by sequence {rp[n]}, so the generators (t) must form a Riesz basis
of L2. In other words, there exist two constants A>0 and B<, such that:
A r[ n]

2
2

m 1

z r [n]

=
p 0 n

p (t

B r[ n]

nT )

2
2

m 1

Where, r[ n] 2 = r [ n] 2 , is L2 norm.
p
2
2
=
p 0 nz

The above-mentioned subspace V() is a single space, the more interesting aspect we are considering
is that x(t) lies in a union of subspaces V p ( ) (0pm-1) x (t ) Vp ( )
In Fourier domain, (4) can be represented as follow:
m 1

X ( ) = R p ( ) p ( )

(5)

p =0

Where, RP() is the discrete-time Fourier transform of rp[n], p() is the Fourier transform of p(t).
We can obtain the DTFT of the i-th channel samples yi [ n ] by (3) and (5):
Yi ( ) =

1
T
1
T

m 1

( 2 n / T ) p ( 2 n / T ) e

n = p = 0
m 1

(6)

R ( )
p

p= 0

j 2 ni / T

( 2 n / T )e

j 2 ni / T

n =

Where, the fact that the R p ( ) is 2 -periodic.

An appropriate matrix represent of (6) is given by:
Y ( ) = H ( ) R ( )

(7)

Where, Y ( ) = (Y0 ( ), Y1 ( ), Ys 1 ( )) '

R ( ) = ( R0 ( ), R1 ( ), Rm 1 ( )) '

We can recover the sampled signal via find the unique solution of the equation. But, this method needs
to use a digital correction filter, which is difficult in achievement in digital electrocircuit. So, we use
reconstruction functions { g p (t nT ) : p 0,1 , m 1} to realize the reconstruction directly without

digital correction filter H ( ) .

3 Reconstruction Algorithms

So x^(t)can be obtained by Fig. 2:

s 1

x (t ) = yi [ n]gi (t nT )
=i 0 nz

(8)

648

Luo name
Junyi et
al. / Energy
Procedia
16 (2012)
645 650
Author
/ Energy
Procedia
00 (2011)
000000

Fig 2.proposed periodic non-uniform sampling

Let

Gi (
=
)e
jt

We can have
s 1

z g (t nT )e

j n

e jt Gi () =
e j 2 ni / T
i= 0

n =

(9)

s 1 +

g (t nT )e
i

i = 0 n =

j (n 2 ni / T )

(10)

We suppose that the signal X()[a, b] is consisted of N frequency bands, each of length no large
than B. dividing the frequency interval [a, b] into s sections and each of equal length B, c0, c1, cs-1 are
utilized to denote the s sections. We define 1/T=B, which can insure that the aliasing will not be occurred
on the inner of each of sections when c0, c1, cs-1 is moved to c0.
Our aim is to find gi (t )(0 i s 1) but not r[ n ] , we can obtain x(t) from inverse Fourier
transformation.
x(t) =
=
=

1
X()ejt d
2 I

1
X0 ()ejt d + X1()ejt d +"+ Xs1()ejt d

2 c0
c1
cs1

(11)

1 s1
X(+lc ()k)ej(+l()k)t d
2 c=0 S0

Substituting (12) into (3)

s 1 +

1
x(t )
=
2

s 1

X ( + l ( )k )
c

c = 0 c0

g (t nT )e

s 1

s 1

s 1

i
n =
+
j 2 ni / T

G ( ) e

c = 0 c0

i =0

j 2 ni / T

n =

Y ( + l ( )k ) g (t nT )e
i= 0

j (( + lc ( ) k ) n 2 ni / T )

G ( + l ( )k ) e
i= 0

1
=
2

i = 0 n =
s 1

j ( + lc ( ) k ) n

jlc ( ) kt

(13)

n =

Combining (13) and (8)

s 1

G ( ) e
i= 0

n =

j 2 ni / T

s 1

e
c= 0

jlc ( ) kt

s 1 +

jlc ( ) kn

(14)

c = 0 n =

An appropriate matrix represent of (14) is given by:

H ( )G ( ) = E ( )

Where,

(15)

649

LuoAuthor
Junyi name
et al. // Energy
650
Energy Procedia
Procedia 16
00 (2012)
(2011) 645
000000

G ( ) = [ G0 ( ), G1 ( ), , Gs 1 ( ) ]

'

h0,0
H ( ) =
hs 1,0

h0,1

hs 1,1

hi ,c ( ) = e jlc ( ) kt

...

h0, s 1

hs 1, s 1

j 2 ni / T

n =

E ( ) = [ e0 ( ), e1 ( ), , es 1 ( ) ]

'

The

Gi ( )(0 i

s 1) can be obtained
G ( )

= H ( ) E ( )

(16)

From (16), the reconstruction functions can be obtained as follow:

1
2
1
=
2

=
gi (t )

z g (t nT )e

G ( )e

j t

j n

d
(17)

4 Simulations

In the section, we will validate the reconstruction algorithm in MATLAB. We design a sampling
system that the sampling channels are s=3. The corresponding nonuniform sample sequences in Fig.2 are
a0 (=
t ) (t nT ) a1 (t ) = (t nT ) a2 (t ) = (t nT 2 )
We define = T / 3 that is the sequence separation between two interleaved uniform sample
sequences. The generate functions 0 (t ) and 1 (t ) are given as follow:

t j 2 30T t
T
t
1 (t ) = sinc( )
T
t j 2 30T t
2 (t ) = sinc( )e
T
0 (t ) = sinc( )e

(18)
(19)
(20)

Where, T is the sampling period.

We suppose that the input multi-band signal
x (t ) = sinc(Bt )( cos(2 t / T ) +1) /
(21)
Where, B is the bandwidth of the input signal. The sampling period T=1/B. Fig 3 shows the conclusion
of simulation. The sampled signal Yi ( ) (0 i s -1) is aliasing as shown in Fig.3a. Fig.3b shows signal in
the third channel as an example to prove that each channel can reconstruct one part of whole signals.
Fig.3c and Fig.3d shows the reconstructed signals and the original signals respectively. Obviously, we
can observe that the signals can be completely recovered by use the proposed reconstruction method in
this paper.

650

Luo name
Junyi et
al. / Energy
Procedia
16 (2012)
645 650
Author
/ Energy
Procedia
00 (2011)
000000
Frequency content of y

Frequency content of X

10

0
0

-1
-10

-2

-20

-4

db

dB

-3

-30

-5
-6

-40

-7
-50

-8
-9

6
frequency (Hz)

10

-60

12

Frequency content of X

10

3
4
frequency (Hz)

x 10

7
7

x 10

Frequency content of X

10

-10

-10

-20
db

db

-20

-30

-30

-40
-40

-50

-50

-60
-70

3
4
frequency (Hz)

-60

x 10

3
4
frequency (Hz)

7
7

x 10

Fig.3 (a) the spectrum of Yi ( ) ; (b) the signal spectrum in the third channel ;(c) The spectrum of reconstructed signals (d) The
spectrum of the original signals

5 Conclusions

In this paper, we use a general framework to treat sampling of multi-band signal. Our interest is that
focused on how to reconstruct signal completely. The approach we chosen are that project the signal over
basis functions and then sample the basis coefficients. The latter focuses on using reconstruction function
to obtain signals. Finally, the simulation proved the method we proposed is feasible.
References
[1] C. H. Tseng and S. C. Chou, Direct downconversion of multiple RF signals using bandpass sampling, in Proc. ICC, May
2003, vol. 3, pp. 20032007.
[2] R. G. Vaughan, N. L. Scott, and D. R. White, The theory of bandpass sampling, IEEE Trans. Signal Processing, vol. 39, pp.
19731984, Sept.1991.
[3] D. M. Akos and M. Stockmaster, Direct bandpass sampling of multipledistinct RF signals, IEEE Trans. Commun., vol. 47,
pp. 983988, July 1999.
[4] G. Hill, The benefits of undersampling, Electron. Design, pp. 6979, July 1994.
[5] Y. C. Eldar, Sampling and reconstruction in arbitrary spaces and oblique dual frame vectors, J. Fourier Analys. Appl., vol.
1, no. 9, pp. 7796, Jan. 2003.