You are on page 1of 4

Calculation of Cycle Lengths In Higher-Order

MASH DDSMs with Constant Inputs


Brian Fitzgibbon and Michael Peter Kennedy

Department of Microelectronic Engineering and Tyndall National Institute


University College Cork, Cork, Ireland
Email: brian.fitzgibbon@tyndall.ie.peter.kennedy@ucc.ie
Abstract-A mathematical analysis is performed to investigate
the periodic behavior of Multi stAge noise SHaping (MASH) Dig
ital Delta-Sigma Modulators (DDSMs). The analysis is performed
on fourth- and fifth-order MASH DDSMs with an odd initial
condition on the first stage and all other states initially zeroed.
We prove that the maximum cycle length for the fourth- and
N+2
fifth-order MASH DDSM is 2
where N is the wordlength

conditions. They showed that an odd initial condition in the


first stage of an EFM or MASH DDSM maximises the cycle
length. This result was proven in [8] for second- and third
order MASH DDSMs. In this work, we complete the proof
for fourth- and fifth-order MASH DDSMs.

of the modulator. In the case of fourth-order modulators, the


maximum cycle length can be achieved when odd digital inputs
are
plied. In the case of fifth-order modulators, the cycle length
is 2

+2 for all digital in uts.


p

I. INTRODUCTION
A DDSM is a discrete-time system whose function is to re
quantize a discrete-amplitude input signal coarsely to produce
a lower resolution output signal. This requantization process
takes place within a negative feedback loop such that the
power of the resulting quantization noise is suppressed within
some signal band of interest [1]. Ideally the quantization noise
introduced by the DDSM is white and uncorrelated with the
DDSM's input sequence. In practice, however, the quantization
error often forms short and repeating patterns, giving rise to
spurious tones in the output spectrum.
Two classes of techniques have been developed to whiten
the quantization noise: stochastic and deterministic. Stochastic
techniques include the use of LSB dithering [2]-[4] and time
varying noise transfer functions [5]. Their goal is to make the
quantization noise asymptotically white and indpendent of the
modulator's input, thereby eliminating spurious tones in the
output spectrum. Deterministic techniques include the setting
of predefined initial conditions [6]-[8], using prime modulus
quantizers [9] and architectural modification [10], [11]. The
goal of these methods is to maximize the cycle length of
the quantization error signal, thereby causing the quantization
power per tone to be minimized.
In this work, we focus on a deterministic technique, namely
setting predefined initial conditions. It has been proven that an
irrational initial condition imposed on the first accumulator of
a third or higher order MASH DDSM driven by a rational
DC input guarantees a spur-free output spectrum [6]. In the
case of a fixed-point digital implementation, the authors of [6]
suggested the use of an odd number as an approximation to
an irrational initial condition. The authors of [7] performed
extensive simulations on various orders of MASH DDSMs
and error feedback modulators (EFMs) to extract an empir
ical design methodology based on setting predefined initial

978-1-4244-8157-6/10/$26.00 2010 IEEE

479

II. MASH DDSM ARCHITECTURE


The basic building block of the MASH DDSM is the first
order error feedback modulator (EFMI) shown in Fig. 1. The
input to the modulator is a digital word with N bits. When
v[n] is greater thanM (2N), the quantizer overflows and the
output signal y[n] will be 1. On the other hand, when v[n] is
less thanM, the quantizer does not overflow and y[n] will be
o. Mathematically, we write:

y[n]

{ 1,

0,

v[n] <M,
v[n] 2:M.

(1)

x[n]
N

Fig. I.

Block diagram of a first-order error feedback modulator (EFMI).

Figure 2 shows a block diagram of an [th order MASH


DDSM comprising a cascade of N-bit EFMl blocks and a
noise cancellation network. In this structure, the negative of
the quantization error from each stage (-edn]) is fed to the
next stage and the output of each stage (ydn]) is fed to the
noise cancellation network, which eliminates the intermediate
quantization noise terms [12]. The output of the [th order
MASH DDSM can be expressed in the Z-domain as:

where X(z) and Ez(z) are the Z-transforms of the input and
the quantizer error introduced by the [th stage. We can write
the negative of the quantization error from the [th stage in

ICECS 2010

Noise Cancellation Network

where

a
b

4 n+10

(8)

6n2 +30n+35

(9)

C =

4 n3 +30n2 +70n+50

3n+6

e = 3n2 +12n+11.

x[nJ
Fig. 2. Block diagram of an zth order MASH DDSM incorporating a cascade
of EFMls

terms of the input and initial conditions as [8]


kl-1

-ez[n]

Iz+

kl_1=O

IZ_ 1 +

k2

+. . . +

kl-2=O

L h +(k1 + I)X )

k1=O

modM,

(3)

22m+(3n+6)2m+3n2 +12n+11,

(4)
III.

FOURTH-ORDER

f(k+1,1)

22k+2 +92k+ 1 +26

422k +182k +26

( t f: f:

k3=O k2=O kl=O

h +(k1 + I)X

) modM

which is divisible by 3. Next, we assume that f( h, k) is


divisible by 3 for some positive integers hand k, i.e

31(22h+(3k+6)2h+3k2 +12k+11).
f(h, k+1)

(16)

22h+(3(k+1)+6)2h+3(k+1)2
+12(k+l)+11

22h+(3k+9))2h+3(k2 +2k+1)
+12(k+l)+11

(5)

22h+(3k+6)2h+3k2 +12k+11

+322h+3(2k+l)+12,

(n+1)(n+2)(n+3)(n+4 )
X
24
(n+l)(n+2)(n+3)
+
I1 modM.
6
=

(15)

Simplifying this equation yields

-e4[n]

22 k +9 . 2k +26+322k +92k ,

Examining f( h, k+1), we obtain

MASH DDSM

Expanding (4) for a fourth-order MASH DDSM, we can


write

-e4 [n]

(14)

where xly denotes that x divides y. Examining f(k + 1, 1),


we obtain

modM

(13)

where N4 = 2m. Consider the case f(l, 1) = 4 8, which is


divisible by 3. We assume that f(k, 1) is divisible by 3 for
some positive integer k, i.e.

31(22k +92k +26),

where Iz denotes the initial condition imposed on the register


of the lth stage and X is the constant DC input. In this work,
we focus on MASH DDSMs where all initial conditions except
for the first stage are set to zero. Consequently, we can rewrite
(3) as

(12)

We examine the periodicity of (7) with the condition that


11 is odd. We show using two-dimensional induction that
the numerator of the second term is divisible by 3 and
consequently that the minimum length solution for N4 must
be 2M to ensure that the second term is an integer. We rewrite
the numerator as

f(m, n)

IZ_
2

(10)
(11)

(6)

In order to find the period N4 of the fourth-order MASH


DDSM, we impose the condition -ern] = -ern + N4 ] on
Eq. (6), obtaining

x
( Nl+aNl+bN4+c
24
Nl+dN4 +e
+
I1 ) - 0 modM,
6
N4

(7)

480

(17)

which is divisible by 3. It follows that f(m, n) is divisible


by 3 for all positive integers m and n, and consequently that
the minimum length solution for N4 is 2M. Using a similar
argument, it can be shown that the numerator of the first term
in (7) is divisible by 6. Consequently, when the input X is
odd, the minimum length solution for N4 such that the first
term is an integer is 4M.
The minimum and maximum cycle lengths for an 18-bit
fourth order MASH DDSM are illustrated using the autocor
relation function in Figs. 3 and 4, respectively. In the case of
a periodic signal, the autocorrelation function contains peaks
that are separated by the period of the signal. The cycle length
is determined by the index with the same autocorrelation value
as at index o. When the input is even, a cycle length of

2 19 is achieved with an odd initial condition, as illustrated


in Fig. 3. When the input is odd, a cycle length of 220 is
achieved, as illustrated in Fig. 4. The effect on the DDSM
output spectrum of using an odd initial condition is illustrated
in Fig. 5. When the initial condition is zero, the spectrum
consists of discrete tones which deviate significantly from the
ideal curve of additive shaped white quantization noise. Using
an odd initial condition maximises the cycle length and the
spectrum approaches the ideal curve.

Power Spectral Density


100 .-----------

Q)

g.
a;
"0

-;; -100
:0
>-

l!'
'l':

rr

0.1 .---------,

-200

-300

U."'_'I"f

1111

-400 '-----.......
-3
10

0.08

Normalized Frequency (Xl[ rad/sample)

Fig. 5. Output spectrum of an 18-bit fourth-order MASH 1-1-1-1 DDSM


with X = 1024 for three different cases: (i) ideal shaped white quantization
noise, (ii) odd initial condition on the first stage (h = 1), and (iii) zero initial
condition on the first stage (h = 0). A Hanning window with 220 terms was
used when computing the spectrum.

0.06
c
o

<3
"

'"

0.04

Simplifying this equation yields


.
________----'
-0.04 L..._
o
-1048576 -524288
524288
1048576

-e5[n]

Lag Index

Fig. 3. Autocorrelation function of the quantization error in an 18-bit fourth


order MASH 1-1-1-1 DDSM with X = 131072 and an odd initial condition
on the first stage (h = 1).

+3)(n+4)(n+5)
( (n+l)(n+2)(n120
X
(n+l)(n+2)(n+3)(n+4) )
+
I1 modM.
24
=

(19)

In order to find the period N5 of the fifth-order MASH DDSM,


we impose the condition -ern]
-ern + N5] on Eq. (19),
obtaining
=

( Nt+aNt +120bNg +cN5+d X


Nt +eNg : fN5+
) 0 modM,
+
2

0.1

N5

0.08

gh

0.06
c

<3
"
0

0
S

0.04

(20)

where

0.02
0

==

b
c

-0.02
-0.04

d
e

-0.06
-1048576

-524288

524288

1048576

Lag Index

Fig. 4. Autocorrelation function of the quantization error in an 18-bit fourth


order MASH 1-1-1-1 DDSM with X = 131073 and an odd initial condition
on the first stage (h = 1).

IV.

FIFTH-ORDER

MASH DDSM

Expanding (4) for a fifth-order MASH DDSM, we can write


modM
(18)

481

5n+15

(21)

10n2 +60n+85
10n3 +90n2 +255n+225
4
5n +60n3 +255n2 +450n+274
4n+20

(22)
(23)
(24)
(25)

6n2 +30n+35

(26)

4n3 +30n2 +70n+50.

(27)

It can be shown by means of two dimensional induction that


the numerator of the second term in (20) is divisible by
6. Consequently, when the initial condition h is odd, the
minimum length solution for N5 such that the second term
in (20) is an integer is 4M. This solution for N5 is also
sufficient to ensure that the first term is an integer, as the
expression Nt + aNt + bNg + cN5+ d is divisible by 30.
The theoretically derived value for the cycle length with an
odd initial condition is verified in Figs. 7 and 8 for an 18bit MASH 1-1-1-1-1 with odd and even inputs, respectively.
Note that the cycle length in both cases is 220, as expected.

The effect on the DDSM output spectrum of using an odd


initial condition is illustrated in Fig. 6. Note that maximizing
the cycle length via the use of an odd initial condition causes
the spectrum to approach that of the ideal curve.

0.1
0.08
0.06
c

0
<)

Power Spectral Density


1 00 r-----------

Qi'
Ci
E

:s

-50

1j

-1 00

.......... .:::
"
.
. ..... .

-0.02
-0.04
-0.06

>-

Q)

0.02

0
:;
<{

50

0.04

-1048576

-1 50

-524288

-200

-250

({

-300

524288

1048576

Lag Index

"

Fig. 8. Autocorrelation function of the quantization error in an 18-bit fifth


order MASH 1-1-1-1-1 DDSM with X = 2 and an odd initial condition on
the first stage (lI = 1).

11

------------
-350 --------------3
10

Normalized Frequency (Xl[ rad/sample)

Fig. 6. Output spectrum of an 18-bit fifth-order MASH 1-1-1-1-1 DDSM


with X = 32768 for three different cases: (i) ideal shaped white quantization
noise, (ii) odd initial condition on the first stage (lI = 1), and (iii) zero initial
condition on the first stage (II = 0). A Hanning window with 220 terms was
used when computing the spectrum.

TABLE I
CYCLE LENGTHS WITH PREFERRED INITIAL CONDITIONS FOR EFMs
WITH WORD LENGTH N (7)

MASH

Minimum

Maximum

Initial

Order m

Cycle Length
2N+1

Cycle Length
2N+2

Condition

4
5

2N+2

2N+2

odd
odd

0.1

0.08

REFERENCES
c

0
<)

0
:;
<{

0.06

0.04

0.02

-0.02 L-___________----'
1048576
-1048576 -524288
o
524288

Lag Index

Fig. 7. Autocorrelation function of the quantization error in an 18-bit fifth


order MASH 1-1-1-1-1 DDSM with X = 1 and an odd initial condition on
the first stage (lI = 1).

V. CONCLUSION
In this work, we have verified mathematically the empirical
results presented in [7] for fourth- and fifth-order MASH
DDSMs. For an odd initial condition on the first accumulator
of the fourth- and fifth-order modulators, the minimum and
maximum cycle lengths are given in Table I, where N is the
wordlength of the modulator. In the case of the fourth-order
MASH DDSM, the maximum cycle length is achieved for
odd inputs. In the case of the fifth-order MASH DDSM the
maximum cycle length is achieved for all inputs.
ACKNOWLEDG MENT

This work is funded in part by SF! under Grant 08/1N.l/1854


and by IRCSET under the Embark Initiative.

482

[I) R. Schreier and G.C. Ternes, Understanding Delta-Sigma Data Convert


ers. New York: Wiley, 2005.
(2) S. Pamarti et a!., "Statistics of the Quantization Noise in I-Bit Dithered
Single-Quantizer Digital Delta-Sigma Modulators," IEEE Trans. Circuits
Syst. I, Reg. Papers, vol. 54, no. 3, pp. 492-503, Mar. 2007.
(3) S. Pamarti and I. Galton, "LSB Dithering in MASH Delta-Sigma D/A
Converters," IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 54, no. 4, pp.
779-790, Apr. 2007.
[4) VR. Gonzalez-Diaz et a!., "Efficient Dithering in MASH Sigma-Delta
Modulators for Fractional Frequency Synthesizers," IEEE Trans. Circuits
Syst. I, Reg. Papers, vol. 57, no. 9, pp. 2394-2403, Sep. 2010.
[5) F. Maloberti et a!., "Time variant digital sigma-delta modulator for
fractional-N frequency synthesizers," in IEEE Int. Symp. Radio-Frequency
Integration Technology, Dec. 2009, pp. 111-114.
[6) M. Kozak and I. Kale, "Rigorous analysis of delta-sigma modulators for
fractional-N PLL frequency synthesis," IEEE Trans. Circuits Syst. I, Reg.
Papers, vol. 51, no. 6, pp. 1148-1162, Jun. 2004.
[7) M.J. Borkowski et a!., "A practical b. modulator design method based
on periodical behavior analysis," IEEE Trans. Circuits Syst. II, Exp.
Briefs, vol. 52, no. 10, pp. 626-630, Oct. 2005.
[8) K. Hosseini and M.P. Kennedy, "Mathematical Analysis of Digital MASH
Delta-Sigma Modulators for Fractional-N Frequency Synthesizers," in
P h.D. Research Microelectronics and Electronics, Jun. 2006, pp. 309312.
(9) K. Hosseini and M.P. Kennedy, "Mathematical Analysis of a Prime
Modulus Quantizer MASH Digital Delta-Sigma Modulator," IEEE Trans.
Circuits Syst. II, Exp. Briefs, vol. 54, no. 12, pp. 1105-1109, Dec. 2007.
(10) K. Hosseini and M.P. Kennedy, "Maximum Sequence Length MASH
Digital Delta-Sigma Modulators," IEEE Trans. Circuits Syst. I, Reg.
Papers, vol. 54, no. 12, pp. 2628-2638, Dec. 2007.
[II) J. Song and I.-C. Park, "Spur-Free MASH Delta-Sigma Modulation,"
IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 57, no. 9, pp. 2426-2437,
Sep. 2010.
[I 2) K. Uchimura et a!., "Oversampling A-to-D and D-to-A converters
with multistage noise shaping modulators," IEEE Trans. Acoust., Speech,
Signal P rocess., vol. 21, no. 12, pp. 1746-1756, Dec. 1991.

You might also like