Anda di halaman 1dari 10

918 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: ANALOG AND DIGITAL SIGNAL PROCESSING, VOL. 50, NO.

SING, VOL. 50, NO. 12, DECEMBER 2003

1.5-V MOS Translinear Loops With Improved


Dynamic Range and Their Applications to
Current-Mode Signal Processing
Carlos A. De La Cruz-Blas, Antonio López-Martín, and Alfonso Carlosena

Abstract—Novel MOS translinear loop topologies for very region of transistors, thus, making it an interesting alternative
low-voltage applications are presented. The inclusion of dc level for nonlinear and linear current–mode signal processing.
shifting, together with a novel biasing scheme based on two Unfortunately, most of the proposed TL loop implementations
MOS transistors in the triode region, allows the operation of the
loops at supply voltages as low as +2 sat maintaining
are not able to exploit the potential of the TL principle for very
low supply voltage [2], [3].
at the same time a large dynamic range. Several current-mode
translinear circuits, both static and dynamic, i.e., geometric mean, To overcome this restriction, we propose in this paper novel
squarer/divider, multiplier and square-root-domain filters, are TL loop topologies able to operate at supply voltages as low as
implemented following this approach, demonstrating on silicon
, and featuring at the same time a large dynamic
the proposed techniques.
range. They are based on a novel loop biasing technique, and
Index Terms—Companding, current-mode circuits, MOS the inclusion of dc level shifters. In the literature, level-shifting
translinear loops, very low-voltage.
techniques have been formerly employed to decrease the supply
voltage of current mirrors [4]–[6]; their versatility has been
I. INTRODUCTION demonstrated experimentally in many practical designs, al-
lowing the artificial reduction of the effective threshold voltage
A NALOG DESIGN is facing a continuous and profound
challenge to satisfy portable electronic equipment needs.
This fact and the down scaling in modern submicron CMOS
of the CMOS process [1]. This strategy was subsequently
extended and applied to more elaborated circuits. For instance,
in [7] and [8], an operational amplifier and a linear transcon-
fabrication processes are forcing to pay attention to low-voltage
ductor employing this technique are proposed, respectively.
analog circuits. Tighter low-voltage requirements confront
In this paper, level-shifting techniques are integrated as a part
designers with the search for new techniques, to revisit current
of TL loops design, in both stacked and up–down topologies,
methodologies and/or to find versatile and new ones at all
leading to a large dynamic range at low supply voltages in both
design levels; at device level, using transistors driven differently
structures. Moreover, a novel biasing scheme employing tran-
(using the bulk terminal) or operating in alternative regions
sistors in the triode region for up–down TL loops is provided,
(weak and moderate inversion); at system level, employing new
achieving these benefits to a larger extent. A complete family
structures and methodologies [1]. For instance, current–mode
of novel low-voltage static and dynamic current-mode circuits
processing exploits the signal-handling capability of currents,
based on the techniques proposed has been built in a 0.8- m
whereas voltages are nonlinearly related to circuit currents
CMOS process, including geometric mean, and squarer/di-
according to their inherent compression law , and
vider circuits, a multiplier/divider and first- and second-order
for MOS and bipolar junction transistors (BJTs), respectively.
square-root-domain filters. Measurement results are provided
These compressed voltages are simultaneously processed along
confirming the versatility of the proposed approach.
with their respective currents, but they play a secondary role in
The paper is organized as follows. In Section II, the TL prin-
the designs. In this way, the signal dynamic range becomes less
ciple is reviewed; Section III discusses the inclusion of floating
influenced by a reduction in supply voltage. Translinear (TL)
voltage sources in MOS TL loops in conjunction with a novel
loops constructed with MOS transistors operating in strong
biasing scheme for the up–down topology. Section IV describes
inversion and saturation fall into this category. The TL principle
the practical implementation of the level shifters. Second-order
is based on large-signal behavior, exploiting a wide operating
effects induced by the level shifters are investigated in Section V.
Simulation and measurement results demonstrating these ideas
Manuscript received March 25, 2003; revised July 11, 2003. This work
are offered in Section VI.
was funded in part by the Spanish Dirección General de Investigación under
Grant TIC2000–0615-C02–02 C. A., and in part by a Scholarship from
CONACyT/México under Contract 118658. This paper was recommended by
Associate Editor F. Duque-Carrillo. II. MOS TRANSLINEAR PRINCIPLE
The authors are with the Department of Electrical and Electronic Engi-
neering, Public University of Navarra, Campus Arrosadía, Pamplona E-31006 The term “translinear” was first introduced by Gilbert [9] in
Spain (e-mail: carlos.aristoteles@unavarra.es; antonio.lopez@unavarra.es;
carlosen@unavarra.es). 1975, emphasizing the fact that trans conductance ( ) is linear
Digital Object Identifier 10.1109/TCSII.2003.820230 with current, for the case of BJT transistor and MOS transistor
1057-7130/03$17.00 © 2003 IEEE
DE LA CRUZ-BLAS et al.: 1.5-V MOS TL LOOPS WITH IMPROVED DYNAMIC RANGE 919

in weak inversion. However, when transconductance is linear


with voltage instead of current like (1)

(1)

(where is a scaling factor), it leads to another class


of translinear circuits that can be implemented by the
quasi-quadratic law of the MOS transistor, i.e., solving (1) for
in terms of and subsequently integrating
Fig. 1. TL loops with voltage sources.
(2)
to operate in saturation. In Section III, it will be shown how the
being an arbitrary constant. This fact is considered as an ex- insertion of floating sources leads to a drastic reduction in the re-
tension of the translinear principle, known as MOS (or voltage) quired voltage for the stacked case and increases dynamic range
translinear principle [2]. The quadratic expression (2) could be for the up–down case thanks to a novel biasing scheme.
achieved using MOS transistors in strong inversion and satura-
tion region, whose first order mathematical model is given by III. ADDING FLOATING SOURCES IN THE TL LOOP

(3) In this section, the effects of including floating voltage


sources1 in MOS translinear loops are investigated. The main
where , , , , and are the gate–source voltage, MOS goals are: 1) to find the appropriate floating voltage source
transconductance parameter, threshold voltage, channel width, values to reduce the supply voltage and 2) to extend the dy-
and channel length, respectively. Comparing (2) and (3), we get namic range of the signals. These characteristics are related to
, and . one each other, since there is usually a tradeoff between them.
According to expression (3), the MOS translinear principle The inclusion of the voltage sources (see Fig. 1) modifies the
was stated as follows: in a loop with an even number of gate- loop (4) as follows:
source connections and with the same number of transistor ar-
ranged clockwise (CW) and counterclockwise (CCW), if it is (5)
assumed that all the transistors operate in strong inversion and
saturation region then where all the transistors have been considered to have the same
ratio. Note that for each transistor a voltage source is
(4) added at its gate, in order to preserve the symmetry, even though
the total effect can be modeled by only one source.
To achieve the same current relationship as in the conven-
Another interesting characteristic, arising from (4) is that the tional TL circuits, it is necessary that (4) and (5) become iden-
resulting relationship between drain currents ideally does not tical, that is
depend on temperature or process variations.
A limitation of this and other techniques based on the (6)
MOS square law is that mobility reduction and velocity
saturation effects in modern deep submicron technologies Observe that no particular restrictions apply on , 2, 3,
notably degrade the square law performance of small devices. 4. Thus, could be either linear or nonlinear, time dependent
Such short-channel effects cause that the transconductance or not. The only condition is to satisfy (6). In addition, it will be
parameter and the threshold voltage of (3) be no longer required that no current flows through the floating sources, in
constant, showing dependences on and [10]. To order to facilitate the proper current injection into the loop tran-
mitigate such effects, larger than the minimum length of the sistors. With this in mind it is possible to determine the correct
fabrication process is required in the transistors that build the values of to accomplish our objectives, i.e., preserving the
TL loops. This choice also reduces mismatching and channel correct operation of the TL transistor loop at low supply volt-
length modulation effects, but a decrement in bandwidth is also ages.
observed. A paper that addresses other second-order effects in As will be demonstrated later, the most evident benefit of em-
MOS TL circuits is for instance [2]. ploying the voltage sources in the stacked loop is the possibility
Essentially, there are two practical translinear topologies for to achieve a correct functionality with supply voltages as low as
the case of four transistors: stacked and up–down [2]. Fig. 1 . For the up–down loop, a novel biasing tech-
depicts both possibilities where, for the moment, the voltage nique will be presented allowing to decrease supply voltage to
sources will be assumed equal to zero. the same value. Since practical implementation of the floating
The up–down topology is inherently better suited for low voltage sources requires this supply voltage, they do not lead to
supply voltages than the stacked one, due to the avoidance of a further reduction. However, they increase the dynamic range
stacked gate-source connections [11], [12] that allows opera- of the TL circuit for this supply voltage level.
tion at , instead of of the stacked 1The terms floating source, voltage source, and level shifter will be used in-
counterpart, where is the minimum drain-source voltage terchangeably.
920 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: ANALOG AND DIGITAL SIGNAL PROCESSING, VOL. 50, NO. 12, DECEMBER 2003

Fig. 2. Stacked configuration, square root cell.

Fig. 4. Novel up–down geometric-mean circuit.

, since voltage at node is in this case. If , it


is possible to design these sources to reduce the supply voltage.
Fig. 3. Simplifying sources. The solution to achieve the sought objective is given by

In order to demonstrate the operation of TL loops including


the floating sources, a geometric-mean circuit will be analyzed. (9)
It will also serve as a basic cell for building other static and
dynamic TL circuits such as squarer/divider, multiplier, as well Observe that the condition keeps in the appro-
as first- and second-order square-root-domain filters. priate operating region, nevertheless, (6) imposes a restriction.
Therefore, has to be forced to to achieve a correct oper-
A. Geometric-Mean Circuit ation of the TL loop. Considering a ,
To simplify the analysis, the effects of will be studied on a voltages out of the range specified by (9) would make that at
specific case, the geometric-mean circuit; however, the general- least one transistor would operate in the triode or cutoff region
ization to other topologies employing the same loop is obvious, (when the upper or lower bound are surpassed, respectively).
as will be seen later. Choosing identical and equal , then the
The geometric-mean circuit is implemented in the stacked voltage at node will be and
configuration, when the from (5) (assuming that (6) is sat- the minimum supply becomes
isfied) are forced to fulfill the conditions below
(10)
(7)
At this point, it is necessary to explain that the possibility to
On other hand, the up–down loop has the following current reach the lowest bound depends strongly on the floating source
requirements: implementation. For instance, assuming that and
, then and . Thus,
(8) only a reduced voltage headroom (lees than ) is available to
design . As will be seen later, a resistor and current mirrors
In both cases, , being for the stacked are the basic elements to implement , demanding a
configuration, whereas for the up–down topology. for operating correctly. This fact will restrict the
supply voltage of the complete circuit.
B. Stacked Configuration
Fig. 2 shows the stacked configuration with the dc level C. Novel Up–Down Configuration
shifters and biasing circuitry included. – constitute the Fig. 4 shows a novel square root cell implemented with the
TL loop and – provide the required biasing currents to up–down translinear loop, where again the simplification of
satisfy (7). sources in Fig. 3 is used. As can be observed, the biasing circuit
Comparing the stacked configuration of Fig. 2 with the gen- is more complex than in the stacked topology [2].
eral scheme in Fig. 1(a), a simplification of the sources ac- Transistors – form the MOS TL loop and the re-
cording to the equivalence shown in Fig. 3 can be made, where maining transistors provide the correct injection of currents
’s have been assumed identical. into the loop in compliance with (8). As mentioned above, the
Assuming for the moment that , the minimum supply biasing circuitry of the up–down topology allows to decrease
voltage needed to properly bias the circuit is determined by the supply voltage to , instead of using the
the branch where flows; thus, this branch requires level shifter to this end at these voltage levels [12]. In this
DE LA CRUZ-BLAS et al.: 1.5-V MOS TL LOOPS WITH IMPROVED DYNAMIC RANGE 921

TABLE I
TRANSISTOR ASPECT RATIOS

sense, a novel alternative is proposed in Fig. 4, where and


are operating in triode region, adding their drain-source
voltages , for setting the proper bias voltage at the Fig. 5. Voltage source implementation.
drain of ( ). Transistors and
are arranged in a negative feedback configuration that sets the ible with the circuit; and 3) no current flowing out or into their
required dc level at the loop transistor sources. When voltage terminals.
at any of the common gate nodes of the loop transistors tends According to previous works, floating sources can be imple-
to decrease, the corresponding gate voltage of either or mented in several ways: 1) MOS transistor in strong or weak
tends to decrease, so that the equivalent resistance implemented inversion [5]; 2) floating-gate MOS transistors [4]; 3) switched
by this transistor in triode region tends to increase. This leads floating capacitor [7], [8]; and 4) resistor with current sources
to an increase in the voltage at the common source node of [8], [13]. Equation (12) imposes a voltage restriction that is dif-
and that compensates for the initial decrease. ficult to achieve with solution 1) because the circuitry needed
This novel biasing scheme allows the avoidance of the usually requires higher supply voltages than .
classical diode-connected MOSFET at the sources of and Nevertheless, in [14] a weak inversion transistor is employed
[2], reducing the requirements from to . to obtain , but it is dependent on threshold voltages and
Thus, the minimum supply level needed to achieve the square technology, so that the operation of TL loop transistors in the
root function is intended region is not guaranteed. Solution 2) is a good alterna-
(11) tive, even though the gate initial charge on the practical design
needs to be removed. Solution 3) requires an extra clock signal,
demanding as condition for biasing the which increments the complexity and the noise. Last, solution
TL transistor loop. This new proposal has advantages over that 4) depicted in Fig. 5, is the simplest and more straightforward
offered in [12] in terms of silicon area and simplicity, while one to implement it, since only current mirrors and resistor
keeping similar dynamic range and bandwidth. Table I shows are required. The process tolerances for can be compensated
the aspect ratios of the MOS transistors in Fig. 4. by tuning . The source value is given by
Once that the novel low-voltage biasing is employed in the
up–down TL loop, the floating voltage sources cannot further (13)
reduce the voltage supply required since their physical imple- where the resistance terminals are high-impedance nodes
mentation also requires a minimum voltage supply of whose voltage is set by the nodes to which the floating source
. However, level shifters can be exploited in this case is attached.
to extend the dynamic range. This is due to the fact that by the The voltage limitation of the proposed source is
inclusion of the level shifters, drain voltages of and for the upper bound, and for the
in Fig. 4 are Volts lower than the corresponding gate-voltage lower one, restricting this fact the lower supply voltage of both
of the TL circuit, so that larger drain-currents can be driven be- translinear loops stacked and up–down to operate correctly.
fore they enter in their triode region. Nonetheless, the maximum
current that can be applied to the loop depends on this fact and V. SECOND-ORDER EFFECTS
the level shifter implementation, i.e., the maximum current will
be reached when either enter in their triode region or The addition of the dc level shifters to the TL loops influ-
level shifters no longer operate correctly. In our TL design, the ences their bandwidth, power consumption and area. A careful
second condition occurs first. The values for , leading to an analysis is needed to minimize the impact of second-order ef-
increment of the dynamic range, are expressed by fects and their influence in the TL loop operation. This Section
describes briefly some important issues and considerations to
(12) reduce these influences.
If (12) is not accomplished, two possibilities arise. For the upper A. Bandwidth
values of (12), at least one of the loop transistors will be forced
Level shifters incorporate extra poles and zeros to the TL
to enter in the triode region wrecking the MOS TL principle. On
loop, and modify the total resistance and capacitance at the
the other hand, lower values of (12) will require larger supply
nodes to which they are connected. To estimate their influence,
voltages for the MOS TL loop to operate correctly.
a small signal analysis was made. In such an analysis, transistor
(see Fig. 4) and floating source were considered. The
IV. PRACTICAL IMPLEMENTATION OF VOLTAGE SOURCES drain-gate connection of increases the parasitic capacitances
The required characteristics for floating sources are: 1) value and the conductance of transistor . These effects can be ne-
bounded according to (9) and (12); 2) supply voltage compat- glected if a proper aspect ratio of the transistors is selected.
922 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: ANALOG AND DIGITAL SIGNAL PROCESSING, VOL. 50, NO. 12, DECEMBER 2003

Thereby the poles and zeros generated by the inclusion of the


floating source have to be faraway from those created by the
transistors of the TL loop. In our design, the simulated small
signal bandwidth of the floating source alone was from 14.7
to 15.5 MHz depending on the tuning current , whereas the
bandwidth of the TL loop alone was approximately 2.5 MHz.
The bandwidth of both circuits (as in Fig. 4) remains practi-
cally 2.5 MHz. The latter data were obtained for a level shifter
with and transistor aspect ratios ,
employing BSIM3 models from a 0.8- m AMS process for the
simulation.
Fig. 6. Stacked: mismatch.
B. Power Consumption and Area
With regard to the power consumption, it is obviously incre- employ the fact that local mismatch of two close elements with
mented. Fig. 5 shows that is distributed to several branches. dimensions , is [15], being , the width
This implies a larger current consumption. For the particular and length of a MOS transistor or a resistor, as the case may be.
case of two floating sources, the static power is given by Thereby, the total variance of could be expressed as

(14)
(17)
where is the total current to implement the floating sources.
To provide an example for , where the first and second term are the contributions of the
and . This represents 23% more than a resistor and mismatch, respectively. is a constant,
geometric mean without floating sources for input currents of whereas depends on the effective voltage ( ) [16].
10 , thereby, the increment of power consumption is notice- The parameters and fully depend on the fabrication
able. process. A realistic value of for a 0.8- process consid-
Area is also an important issue in circuit design. For the par- ering transistor aspect ratios of and resistors of
ticular case of floating sources, low and values are required is 1.5 mV, a 0.2% of a nominal .
for increasing bandwidth and not influencing substantially the To provide realistic figures, some simulations were carried
MOS TL loop behavior. The area of the floating sources was out. Fig. 6 shows simulation results for a geometric mean
8.1 , i.e, a 13% more than for a geometric-mean cir- with a fixed input current A, a sweep for
cuit without floating sources employing an AMS 0.8- m tech- from 5 to 20 A and different values of . Note that due to
nology. the voltage mismatch the curves deviate according to (16). The
quadratic term in (16) can be neglected in this case due to the low
C. Mismatch value of . Considering this analysis, the viability to develop
a geometric-mean cell on silicon using a stacked translinear con-
Another second-order effect is the mismatch of floating figuration with floating sources is demonstrated. Obviously, a
sources. After the simplification of the sources (Fig. 3) the tradeoff in terms of area, power consumption and minimization
up–down topology is free from second-order effects by the of second-order effects exists.
introduction of the voltage sources. This is not the case for the
stacked topology where in spite of the simplification according
VI. SIMULATION AND MEASUREMENT RESULTS
to Fig. 3, two voltage sources remain within the loop. In this
case, due to mismatch in the floating batteries, (5) is modified In general, MOS TL loops constitute versatile building
as follows: blocks with a wide range of applications. Some examples
are vector sum, root-difference squares, RMS-dc converters
(15) [17], [18], differential linear transconductor [19], nonlinear
transconductors to implement square-root-domain filters
(companding techniques) [20] etc. In this section, the results
where is the error voltage caused by mismatching, consid-
obtained for the new TL loop geometric-mean circuit are dis-
ered constant. Rearranging (15) according to the current values
cussed. Then, a squarer/divider cell is designed, demonstrating
in (7), and solving for
that the methodology applied to the geometric-mean circuit
can be readily extended. Once both circuits are implemented,
(16) they can be combined to build more complex circuits, such as
multipliers and square-root-domain filters, which are described
Equation (16) shows that mismatched sources lead to addi- at the end of this Section.
tional nonlinear terms and sensitivity to temperature and fabri- Simulation (BSIM3 models) and measurement results have
cation process through the parameter . The error is mainly been obtained using a 0.8- m DPDM n-well CMOS technology,
due to the mismatched resistor in the floating batteries and with a single supply voltage as low as 1.5 V. Only circuits based
mismatch in the bias current . To estimate , it is feasible to on the novel up–down TL loops were fabricated, due to their
DE LA CRUZ-BLAS et al.: 1.5-V MOS TL LOOPS WITH IMPROVED DYNAMIC RANGE 923

Fig. 7. Stacked simulation results. Fig. 9. Squarer/divider: dc Characteristics.

Fig. 10. Microphotograph of a geometric mean and Squarer/Divider.

B. Squarer/Divider Cell
The geometric-mean cell discussed above in an isolated con-
Fig. 8. Up–down measurement results.
figuration does not offer many design possibilities. To overcome
this drawback it is necessary to implement its dual function, a
squarer/divider, thus expanding their versatility and providing
superior performance. However, simulation results are also pro- flexibility to create more elaborated circuits and systems.
vided for the stacked loop. A squarer/divider is obtained by slightly modi-
fying the geometric mean of Fig. 4, by interchanging input and
A. Geometric Mean output and adapting their input and output impedances [21].
Fig. 9 depicts the dc measurement results of the squarer/di-
1) Stacked Topology: The stacked topology was simulated vider, for , and different values
first at supply voltages of 1.5 V, which would be impossible to of , where is a control current (see Fig. 5) to establish the
achieve without the insertion of floating sources. Simulations value of the floating source . The dashed line ( )
were performed with V and a nominal resistance of demonstrates the improvement in accuracy and dynamic range
. of the circuit achieved by ( ) although the price
Fig. 7 shows different plots over a range of , where it is to be paid by this superior performance is the increase in area
interesting to note the good behavior for values larger than and power consumption.
2.5 . For this supply voltage level, the stacked structure The up–down geometric mean and squarer/divider were fab-
drives some transistors in the ohmic region for very low ricated in the mentioned 0.8 DPDM n-well CMOS process.
currents, leading to the errors observed in Fig. 7 near zero. A microphotograph of the cells is shown in Fig. 10, being their
They are strongly reduced for higher supply voltages and lower total area 0.15 . TL loop transistors are arranged in a
values of for the transistors that build the TL loop. common–centroid topology. Table I shows the transistor aspect
2) Up–Down Topology: As mentioned above, owing to the ratios of both circuits.
novel biasing proposal, the up–down TL loop and level shifter
require in this case a minimum supply voltage of 1.5 V to op- C. Multiplier/Divider
erate correctly. Therefore, level shifters were employed toward
a considerable extension of dynamic range, as can be confirmed A one quadrant multiplier/divider (see Fig. 11) is easily im-
plemented cascading a geometric mean and a squarer/divider
by the following measurement results.
Fig. 8 shows measurement results for with cell [21]. This can be mathematically demonstrated by (18).
, a sweep for from 0 to 25 . The dashed and dotted
line represent for ( ) and (18)
, respectively. Note the increment around 26% for the input
current and the corresponding 9% for the output current , Another multiplier/divider for instance in [22] and [23] can be
evidenving the beneficial influence of the floating sources. obtained employing another algebraic combinations through
924 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: ANALOG AND DIGITAL SIGNAL PROCESSING, VOL. 50, NO. 12, DECEMBER 2003

Fig. 12. Implementation of Cv


_ through up–down squarer/divider and
geometric-mean cells.
Fig. 11. One quadrant multiplier/divider: dc characteristics.
To design the square-root-domain filter, a mapping on (19)
is required. To this end, consider that the output current is
geometric mean and squarer/divider cells as primitives. The
nonlinearly related to a certain voltage by
multiplier/divider implementing (18) was fully designed and
tested in the laboratory providing good agreement with the
(20)
responses obtained in the simulation process.
Experimental results reflecting the static behavior of the one i.e the quadratic behavior of a MOS transistor operating in
quadrant multiplier are shown in Fig. 11, where is depicted strong inversion and saturation region (square-root domain).
for A, A, A and from 0 (lower It was demonstrated in [12] that substituting (20) in (19),
line) to 6 A (upper line) in 1- A steps. Breadboard parasitics applying the chain rule ( ), scaling by a factor
avoid measuring correctly the 3-dB bandwidth. The simulated (a linear capacitance) to translate the equations in the current
one was approximately 2.5 MHz. This relatively poor value is mode and solving for [27], (21) is obtained.
mainly due to the large length of the MOS loop transistors em-
ployed. Total harmonic distortion (THD) measurements were (21)
also made employing an input sinusoid of 1 kHz with dc com-
ponent of 6 A, leading to a THD value lower than 2% for a gain
where the cutoff frequency and dc gain are related to and
factor of approximately 1.7 and amplitude of the input signal as
by
large as 5 A. Obviously, when lower gains [see (18)] are
set, these distortion figures are notably improved.
(22)
D. Square-Root-Domain Filters
Equation (21) represents the dynamic nonlinear large signal
Now first- and second-order filters employing instantaneous behavior of the first-order square-root-domain filter [28]. In
companding techniques will be presented making use of the spite of strong nonlinearities that are involved, the –
novel up–down TL loops proposed here as basic building relationship remains linear, as can be confirmed by the exper-
blocks. The companding paradigm entirely exploits the imental results.
nonlinear characteristics of the devices, firstly compressing Equation (21) can be implemented by the novel up–down
the input voltages, then properly processing, being finally TL loops proposed here. Thus, a MOS companding filter can
expanded at the output. Often, the intermediate step is achieved be designed employing two geometric-mean circuits and one
through nonlinear processing and thus the coined name of: squarer/divider [28]. Firstly, a nonlinear circuit block that imple-
”externally linear internally nonlinear systems” [24], being ments the RHS first term of (21) is made cascading a squarer/di-
companding techniques a subset of those. The nonlinear vider and a geometric mean, where the denominator is injected
blocks are implemented exploiting the dynamic translinear by the current of a MOS transistor (20), whose gate is connected
principle and its implicit current mode operation, maximizing to . The second term is easily achieved by a geometric mean.
the dynamic range available. Depending on the law employed Fig. 12 shows the design strategy, where the blocks represent the
for compression, the system can be defined as log domain [25], basic TL loops offered in the previous sections. In the fabricated
[26] or square-root domain [20] via bipolar (MOS in weak prototype the three cells were integrated, occupying a total area
inversion) or MOS transistors in strong inversion, respectively. of 0.22 . The aspect ratio of MOS transistor to obtain the
1) First-Order Square-Root-Domain Filter: Assume the mapping (20) was . Circuit response was tested
implementation of a current mode integrator expressed by the on the filter with a single supply voltage and an
following; external capacitor . The simulated bandwidth (small
signal) of the cells to build the filter was 2.5 MHz, very similar
(19) to that obtained in the multiplier/divider circuit.
Fig. 13 describes the frequency response of the first-order
being , and the dot the dc gain, cutoff frequency square-root-domain filter for from 2 A to 8 A in
and time differentiation operator, respectively. 2 A steps. It can be appreciated the independent and nonlinear
DE LA CRUZ-BLAS et al.: 1.5-V MOS TL LOOPS WITH IMPROVED DYNAMIC RANGE 925

Fig. 13. Measurements of frequency response of the first-order Fig. 14. Second-order square-root-domain filter: Q-tuning measurements.
square-root-domain filter.
TABLE II
COMPARISON WITH FORMER BIQUAD FILTERS
tuning according to (22). The tuning range of the filter was mea-
sured, being from 2 to 10 kHz. Distortion measurements were
also carried out for a 1-kHz sine input current with dc level of 6
A and the amplitude from 0 to 6 A in 1 A steps. THD less
than 2% can be achieved for amplitudes up to 5 A.
2) Second-Oder Square-Root-Domain Filter: A step
forward leads to the implementation of a second-order
square-root-domain filter by extending the methodology Making use of the mapped state space, (24) and (25) can be
established in the previous section. However, to synthesize expressed in terms of and as follows:
more arbitrary square-root-domain filter or even any class of
companding system, more general methods are discussed in [3] (26)
and [29].
Consider the implementation of a current mode second-order (27)
low-pass filter expressed by the following:
In spite of the strongly nonlinear behavior of (26) and (27), the
input–output relationship remains linear, making a system ex-
(23)
ternally linear but internally nonlinear through a square-root law
for compressing the internal voltages. The representation in (26)
, and the dot being the dc gain, and time differentiation op- and (27) allows to express the parameters of the biquad in terms
erator, respectively. Now defining an auxiliary current of the current square-root ( ) by
(where is an arbitrary constant), (23) can be
decomposed in (24) and (25) (28)

Now that the state equations are in current mode, we can pro-
(24) ceed as for the first-order filter of the previous section to im-
plement (26) and (27). Thereby, three novel squarer/divider and
(25)
four geometric-mean cells as those in Figs. 8 and 9 are required
to build the complete second-order square-root-domain filter.
(24) is obtained substituting (25) in (23) and solving for . A prototype of the filter was fabricated occupying an area
This state variable representation was selected for obtaining a of 0.5 approximately and experimental results were ob-
physically realizable circuit [27]. tained. For instance, Fig. 14 shows the tuning for
In order to implement an instantaneous companding filter A and from 2 to 5 A in steps of 1 A. Obviously,
through the quadratic law of the MOS transistor, a mapped state some deviations (about 10%) respect to similuation results were
space is required. This mapping is achieved making appreciated due to the fabrication process tolerances. The THD
and , whose deriva- figures and the tuning of cutoff frequency were very similar to
tives can be expressed by and the first-order filter case.
, respectively. Voltages and represent the To provide more insight with the technique proposed here,
gate-source voltage of MOS transistor where voltage swings are a comparison was made with formerly reported biquad imple-
compressed by the square-root of and , respectively. mentations. Table II summarizes this comparison, where some
926 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: ANALOG AND DIGITAL SIGNAL PROCESSING, VOL. 50, NO. 12, DECEMBER 2003

important parameters of the filters were considered. Filters [28] [3] A. J. López-Martín and A. Carlosena, “Systematic design of com-
and [3] employ the square-root-domain approach, whereas filter panding systems by component substitution,” Analog Integrated
Circuits Signal Processing, vol. 28, pp. 91–106, 2001.
[8] is a – design. are the aspect ratios of TL loop tran- [4] S. Yan and E. Sánchez-Sinencio, “Low voltage analog circuit design
sistors. THD measurements were made at 1 kHz – 16 techniques: A tutorial,” IECE Trans. Analog Integrated Circuits Syst.,
in this work, 10 kHz – 30 in [3], and 100 kHz, 0.8 vol. E00–A, no. 2, pp. 197–203, February 2000.
[5] J. Ramírez-Angulo, “Current mirrors with low input voltage require-
in [8], respectively. ments for built in current sensor,” in Proc. IEEE Int. Symp. Circuits Sys-
Note that thanks to the inclusion of dc level shifters and the tems, June 1994, pp. 529–532.
novel biasing scheme of the proposed up–down TL loops, a con- [6] S. S. Rajput and S. S. Jamuar, “Low voltage analog circuit design tech-
niques,” IEEE Circuits Syst. Mag., vol. 2, pp. 24–42, Mar. 2002.
siderable saving of supply voltage from in [28] [7] Y. Tand and R. L. Geiger, “A 0.6 V ultra low voltage operational am-
and in [3] to V can be appreciated, plifier,” in Proc. IEEE Int. Symp. Circuits Systems, Scottsdale, AZ, May
preserving similar performance in our design with regard to the 26–29, 2002.
[8] F. Muñoz, A. Torralba, R. G. Carvajal, and J. Ramírez-Angulo, “Two
former biquadratic square-root-domain filters. For the case of new VHF tunable CMOS low–voltage linear transconductor and its ap-
the bandwidth, the filter in [28] offers a very large one. This plication to hf GM–C filter design,” in Proc. IEEE Int. Symp. Circuits
larger bandwidth may be due to the fact that the length of TL Systems, May 2000, pp. 173–176.
[9] B. Gilbert, “Translinear circuits: An historical overview,” Analog Inte-
loop transistors is smaller and the biasing current is also high
grated Circuits Signal Processing, vol. 9, pp. 95–118, 1996.
(200 A), whereas, in our case, the biasing current was 8 A. [10] H. Iwai, “CMOS technology—year 2010 and beyond,” IEEE J. Solid-
As described in [28], the small value of the transistors leads State Circuits, vol. 34, pp. 357–366, Mar. 1999.
to a very small output impedance at the capacitor nodes. There- [11] E. Seevinck, “CMOS translinear circuits,” in Analog Circuit Design,
Most RF Circuits, Sigma-Delta and Translinear Circuits, W. Sansen, R.
fore, as mentioned in [28], strong deviations appear in (26) and J. V. D. Plassche, and J. H. Huijsing, Eds. Norwell, MA: Kluwer, 1996,
(27) causing distortion, and significantly degrading linearity and pp. 323–336.
filter performance. [12] A. J. López-Martín and A. Carlosena, “1.5 V CMOS companding filter,”
Electron. Lett., vol. 38, no. 22, pp. 1299–1300, Oct. 2002.
With regard to the filter in [3], the novel-loop-biasing tech- [13] J. Ramírez-Angulo, R. G. Carvajal, J. Tombs, and A. Torralba, “Simple
nique and the inclusion of level shifters lead to considerable im- technique for opamp continuous-time 1 V supply operation,” Electron.
provements in terms of supply voltage, power consumption, and Lett., vol. 35, pp. 263–264, Feb. 1999.
[14] S. S. Rajput and S. S. Jamuar, “Low voltage, low power, high perfor-
silicon area, beyond the enhancement due to the more modern mance current mirror for portable analogue and mixed mode applica-
fabrication process. tions,” Inst. Elect. Eng. Proc. Circuits, Devices Syst., vol. 148, no. 5, pp.
The comparison of [8] and our design is quite difficult due to 273–278, Oct. 2001.
[15] M. J. M. Pelgrom, A. C. J. Duinmaijer, and A. P. G. Welbers, “Matching
the different approaches. In fact, [8] offers an internally linear properties of MOS transistors,” IEEE J. Solid-State Circuits, vol. 24, pp.
voltage-mode circuit and in our design an internally nonlinear 1433–1440, Oct. 1989.
current-mode processing. The bandwidth of the – is higher, [16] S. J. Lovett, M. Welten, A. Mathewson, and B. Mason, “Optimizing
MOS transistors mistmatch,” IEEE J. Solid-State Circuits, vol. 33, pp.
however, area and power consumption are similar. 147–150, Jan. 1998.
[17] J. Mulder, A. C. V. D. Woerd, W. A. Serdijn, and A. H. M. V. Roermund,
“An RMS–DC converter based on the dynamic translinear principle,”
VII. CONCLUSIONS IEEE J. Solid-State Circuits, vol. 32, pp. 1146–1150, July 1997.
[18] A. J. López-Martín and A. Carlosena, “A 1.5 V current–mode CMOS
The level-shifting technique by means of floating sources in RMS–to–DC converter,” Analog Integrated Circuits Signal Processing,
MOS translinear loops for both stacked and up–down topolo- vol. 36, pp. 137–143, 2003.
gies was investigated. The main goals were a reduction of supply [19] M. H. Eskiyerli, A. J. Payne, and C. Toumazou, “State space synthesis
of integrators based on the MOSFET square law,” Electron. Lett., vol.
voltage or improvement in the dynamic range. It has been shown 32, no. 6, pp. 505–506, Mar. 1996.
that for the stacked case, dramatic savings in supply voltage p
[20] J. Mulder, A. V. D. Woerd, W. Serdijn, and A. V. Roermund, “Cur-
rent-mode companding x-domain integrator,” Electron. Lett., vol. 32,
can be achieved. For the up–down case, thanks to a novel bi-
no. 3, pp. 198–199, Feb. 1996.
asing scheme, the main benefit is reflected on an extended dy- [21] A. J. López-Martín and A. Carlosena, “Current-mode multiplier/divider
namic range. Both topologies are able to work at supply volt- circuits based on the MOS translinear principle,” Analog Integrated Cir-
ages as low as . Second-order effects by the cuits Signal Processing, vol. 28, pp. 265–278, 2001.
[22] K. Tanno, O. Ishizuka, and Z. Tang, “Four–quadrant CMOS cur-
mismatched sources, the impact over the bandwidth and the in- rent–mode multiplier independent of device parameters,” IEEE Trans.
crease in power consumption were analyzed, highlighting the Circuits Syst. II, vol. 47, pp. 473–477, May 2000.
robustness of the up–down loop. Experimental results of static [23] W. Gai, H. Chen, and E. Seevinck, “Quadratic–translinear cMOS multi-
plier–divider circuit,” Electron. Lett., vol. 33, no. 10, pp. 860–861, May
cells based on the proposed up–down TL loop, such as geo- 1997.
metric mean, squarer/divider, multiplier/divider, and dynamic [24] Y. Tsividis, “Externally linear, time-invariant systems and their appli-
systems like first and second-order square-root-domain filters cation to companding signal processors,” IEEE Trans. Circuits Syst. II,
vol. 44, pp. 65–85, Feb. 1997.
were offered providing more insight and validating the proposed [25] E. Seevinck, “Companding current–mode integrator: A new curcuit prin-
strategy. The topologies proposed constitute an interesting alter- ciple for continuous–time monolithic filters,” Electron. Lett., vol. 26, no.
native for very low-voltage analog processing. 24, pp. 2046–2047, Nov. 1990.
[26] D. R. Frey, “Exponential state space filters: A generic current mode de-
sign strategy,” IEEE Trans. Circuits . Syst. I, vol. 43, pp. 34–42, Jan.
REFERENCES 1996.
[27] , “State-space synthesis and analysis of Log-Domain filters,” IEEE
[1] E. Sánchez-Sinencio and A. G. Andreou, Low–Voltage/Low–Power In- Trans. Circuits Syst. II, vol. 45, pp. 1205–1211, Sept. 1998.
tegrated Circuits and Systems. Piscataway, NJ: IEEE Press, 1999. [28] M. Eskiyerli and A. Payne, ““Square Root Domain” filter design and
[2] R. J. Wiegerink, Analysis and Synthesis of MOS Translinear Cir- performance,” Analog Integrated Circuits Signal Processing, vol. 22,
cuits. Norwell, MA: Kluwer, 1993. pp. 231–243, 2000.
DE LA CRUZ-BLAS et al.: 1.5-V MOS TL LOOPS WITH IMPROVED DYNAMIC RANGE 927

[29] C. Psychalinos and S. Vlassis, “A systematic design procedure for Alfonso Carlosena was born in Navarra, Spain, in
Square-Root-Domain circuits based on the signal flow graph approach,” 1962. He received the M.Sc. degree (Hons.) and
IEEE Trans. Circuits Syst. I, vol. 49, pp. 1702–1712, Dec. 2002. the Ph.D. degree in physics from the University
of Zaragoza, Zaragoza, Spain, in 1985 and 1989,
respectively.
From 1986 to 1992, he was an Assistant Professor
Carlos A. De La Cruz Blas was born in Juchitan, in the Department of Electrical Engineering and
Oaxaca, México, in 1975. He received the M.Sc. de- Computer Science at the University of Zaragoza.
gree in electronic engineering from the Instituto Na- In October 1992, he joined the Public University of
cional de Astrofísica Óptica y Electrónica (INAOE), Navarra, Pamplona, Spain, as an Associate Professor
Puebla, México, in 1999. He is currently working to- and where he has also served as Head of the
ward the Ph.D. degree in low–voltage analog inte- Technology Transfer Office. In March 2000, he was promoted to Full Professor
grated circuits at the Public University of Navarra, at the Public University of Navarra. He has also been a Visiting Scholar with
Pamplona, Spain. the Swiss Federal Institute of Technology, Zurich, Germany and New Mexico
State University, Las Cruces. His current research interests include analog
circuits and signal processing, digital signal processing and instrumentation.
On these topics, he has published over 60 papers in international journals and
has a similar number of presentations at various conferences.

Antonio J. López-Martín was born in Pamplona,


Spain, in 1972. He received the M.Sc. and Ph.D.
degrees in electrical engineering from the Public
University of Navarra, Pamplona, Spain, in 1995
and 1999, respectively.
Currently, he is an Assistant Professor with the
Public University of Navarra. He was formerly
a Visiting Professor with the New Mexico State
University, Las Cruces, and an Invited Researcher
with the Swiss Federal Institute of Technology,
Zurich, Switzerland. His research interests include
low-voltage analog and mixed-mode integrated circuits, integrated sensor
interfaces, analog and digital signal processing and communication systems.
He has coauthored over 60 papers on these topics in international journals and
conferences.

Anda mungkin juga menyukai