1. Introduction
Since the introduction of current-mode signal processing, second generation current conveyors (CCII) [
1] have been employed in various circuits, such as inductance simulators [
2,
3], filters [
4,
5,
6], oscillators [
7,
8,
9], and instrumentation amplifiers [
10,
11,
12] as possible substitutes for conventional operational amplifiers (OAs). It is generally known that the OA-based circuits suffer from many disadvantages, mainly low frequency performance resulting from the limited gain-bandwidth product of OAs. Actually, the introduction of CCII has created a valid alternative to OAs in analog circuit design area, and circuits employing CCIIs offer high frequency operation and simplicity. Despite these benefits, the main weakness of CCII is its output signal in the form of current at the high impedance Z port of CCII. Therefore, CCII is not a good block for applications requiring output signal in the form of voltage.
In [
13,
14,
15], the concept of dual topology of CCII, called a second generation voltage conveyor (VCII), was introduced. For this duality, VCII shows the additional advantage of providing output signal in the form of voltage. VCII has a low impedance current input port (Y), a high impedance current output port (X) and a low impedance voltage output port (Z). These impedance levels at Y, X and Z ports are of great importance in both voltage based and current based applications. In [
16], the authors showed that most of the analog signal processing circuits are realizable by VCII with the advantage of providing output voltage signal at the low impedance Z port of VCII. In [
17], a comparison between OA-based circuits and VCII-based circuits was drawn, showing that the problem of the cross-talk effect among inputs in a conventional OA-based non-inverting summing amplifier can be cancelled out using VCII. Literature survey shows the possibility of realizing various VCII-based analog circuits such as filters, integrators, differentiators, rectifiers, and impedance simulators [
16,
17,
18,
19,
20,
21,
22,
23,
24,
25,
26,
27].
As VCII design is a new research area, the VCII implementations reported up to this point lack the maximum efficiency. Analyzing the previously reported VCII circuits, it appears that most of them [
16,
17,
18,
19,
20,
21,
22,
23,
24] are designed in class A, which has limited current drive capability at the X port. Due to high power consumption and poor current drive capability of circuits including class A VCIIs, they are unsuitable for low power applications and off-chip loads [
28], while the class AB VCII with high current drive capability finds wide application as an integrated circuit front-end block. However, designing circuits with high current drive capability under low supply voltage restriction is a very challenging task. For example, to drive 10 mA in 0.18 µm technology with an aspect ratio of 100 µm/0.5 µm, a gate-source voltage equal to 2 V must be provided [
29] for the PMOS transistor, while the maximum allowed supply voltage is 1.8 V. Distortion due to the transistor channel modulation effect (λ effect) is another serious issue [
29] in fine technologies. It is worth mentioning that the conventional cascode structures that were conventionally used to mitigate the channel length modulation effect are not practical, due to the limited allowed supply voltage.
Literature survey shows that there are very few VCII circuits designed in class AB [
25,
26,
27,
30]; however, all of them suffer from poor current drive capability. In [
25], a flipped voltage follower (FVF)-based VCII circuit was presented, which can provide an output current up to ±2 mA. In the VCII circuit reported in [
26], the maximum current at X terminal is limited to ±0.5 mA. The VCII circuit presented in [
27] is optimized for the minimum number of transistors. Its current drive capability is limited to ±1.065 mA. The class AB VCII reported in [
30] is based on the simple translinear principle. Unfortunately, its cur-rent drive capability is only ±1.22 mA.
In this paper we propose a VCII with high current drive capability at the X terminal. The circuit operates in class AB for increasing current swing and reducing power consumption. The current splitting method used previously in current output stages was employed to increase the linearity of output voltage at the Z port and to obtain a high current drive capability at the X port. Due to low voltage restrictions, cascode structures are avoided and the required high linearity in the X port output current is maintained using simple controlling circuits by means of three transistors. The proposed circuit is able to drive a maximum current of ±10 mA at the X port at ±0.9 V supply with a power consumption of 382 µW.
The organization of this paper is as follows. In Section II, details of VCII internal circuit design are given. The existing challenges and design considerations to increase current swing are also discussed in this section. Simulation results are presented in Section III and finally Section IV concludes the paper.
2. Class AB VCII Realization
The internal structure of VCII is composed of a current buffer (CB) and a voltage buffer (VB) as shown in
Figure 1. Y is a low impedance current input port. The input current to the Y port is transferred to a high impedance X port through CB. In practice finite impedance is connected to the X port to convert the conveyed current into the voltage. Then, the voltage produced at the X port is buffered to the low impedance Z port by VB. The matrix operation of a VCII is expressed as:
where α and β are non-ideal VB and CB gains, respectively. Other requirements are high current drive capability in Y and X ports, low THD for the current transferred from Y to X ports, and high voltage swing at the Z port. In all applications reported in [
16,
17,
18,
19,
20,
21,
22,
23,
24,
25,
26], both CB and VB sections are not included in the negative feedback loop, and as a result, overall performance is determined by VCII linearity and port impedances.
In this Section, design considerations of a high drive class AB VCII under low supply voltage are discussed. The analysis is twofold, where the first of which is the CB section.
Figure 2 shows a conventional class AB current buffer [
30]. The translinear loop created by M
1–M
4 sets the value of the input voltage at zero for I
in = 0. Input current is transferred to output node through simple current mirrors (M5-M6, M7-M8). Input impedance is approximately r
in ≅ (gm
2) − 1//(gm
4) − 1. The required minimum supply voltage is:
where Vgs, Vsg, Vds and Vsd are gate-source voltage, source-gate voltage, drain-source voltage, and source-drain voltage, respectively, of the related transistor. In Equation (2), Vsg
M5 and Vgs
M7 are very large for high drain currents. For instance, as mentioned in the introduction, in 0.18 µm CMOS technology, by setting the aspect ratio of M5 at 100 µm/0.5 µm, Vsg
M5 is approximately 2 V for a drain source current of 10 mA, which exceeds the allowed supply voltage of technology [
29]. Evidently, for larger currents, the gate-source voltage is even higher. In the saturation region, the minimum value of the drain-source voltage is Vds(sat) = Vgs – Vth; therefore, in Equation (2), the Vgs terms play a more important role in increasing supply voltage, so if Vgs is replaced with Vds, supply voltage is reduced significantly.
Another problem associated with the conventional class AB current buffer of
Figure 2 is harmonic distortion caused by the channel length modulation effect or the λ effect of M
6 and M
8. In other words, in the upper side, as the drain-source voltage of mirroring transistors M
5 and M
6 are not equal, the current transferred to the output node is not exactly equal to the input current according to:
where µ, C
ox, W
i and L
i (for I = 5–6) are the carrier mobility, gate oxide capacitance, channel width and channel length, respectively, of the related transistor. The same holds in the lower side of the circuit. The problems of large Vgs for M
5 and M
7 along with the nonlinearity caused by the non-zero value of λ are eliminated in the topology shown in
Figure 3, where the minimum required supply voltage is reduced to:
To increase the current drive capability of CB, the required voltage of M
2, M
4, M
5 and M
7 must be reduced, through introduction of M
9 and M
10 splitting transistors [
29]. Transistor M
9 in the upper half has an aspect ratio m times larger than the aspect ratio of M
5. Similarly, in the lower half, M
10 has an aspect ratio m times larger than that of M
7. Therefore, a large part of the input current is provided by splitting transistors M
9 and M
10. As a result, the currents in M
5 and M
7 are reduced and consequently their Vds is reduced, resulting in lower required supply voltage. Therefore, considering Vds(sat) = Vgs − Vth and λ ≅ 0, Equation (5) can also be expressed in terms of drain-source currents as:
or:
with the usual meaning of symbols for
Ki = μCox(Wi/Li) for
i = 2, 4, 5, 7.
From Equation (6a,b), the splitting transistors results in a significant decrease of the supply voltage. In addition, the negative feedback loop provided by the splitting transistors at the input node reduces the Y port impedance by a factor of (1 + m) as follows:
The channel length modulation effect is reduced using two simple controlling circuitries [
31]. In the upper half, the drain-source voltages of M
5 and M
6 are kept equal by a negative feedback loop established by M
C1–M
C2. Similarly, in the lower side of the circuit, the drain-source voltages of M
7 and M
8 are kept equal by M
C3–M
C4. Here, M
L1 and M
L2 are simple level shifters used to provide appropriate bias voltage at drain nodes of M
C2 and M
C4, respectively. Current sources I
B2-I
B7 are used to bias transistors in controlling circuitries. In the lower half, the drain current of M
C3 is fixed by the I
B5 current source. Therefore, its gate-source voltage is a fixed value, and it transfers V
X to the gate of M
C4. Any difference between V
X and the drain voltage of M
7 is compared by M
C4 and the appropriate voltage is produced at its drain node, which is applied to the gate of mirroring transistors M
7–M
8, and M
10 by M
L2. To elaborate the operation of controlling circuitries of the lower half, let us consider the case that V
X is reduced. Due to the channel length modulation effect, a reduction in V
X tends to reduce the M
10 drain current. In this case, the gate voltage of M
C4 is also reduced, which results in an increase in its drain voltage, which is applied to the gate of M
7–M
8, and M
10 through M
L2. As a result, the drain current of M
10 is kept constant regardless of V
X variation.
Similarly, if Vout is increased, the gate voltage of M7–M8, and M10 is reduced to avoid any increase in output current.
Therefore, any variation of the M
10 drain current due to variations in
VX is compensated for. A similar procedure occurs in the upper half by M
C1–M
C2 transistors. As the
λ effects of M
6 and M
8 are compensated for by control circuitry, we can ignore their effect on output impedance and therefore, the impedance at the output node is approximately found as:
To avoid complications, Equation (8) is derived by the assumption that the open loop gain of the λ effect cancelling circuit is infinite. However, in practice, due to its limited gain, it is expected that the impedance at the X port is smaller than Equation (8).
In
Figure 3, in the λ effect cancelling circuit at the upper side, there are four poles located at the source of M
C2, gate of M
C1, source of M
L1 and drain of M
C2. Due to low impedance values at the source of M
C2, gate of M
C1 and source of M
L1, their related poles are smaller than the pole related to the drain of M
C2. To grantee frequency stability, a compensation capacitor can be added to the drain of M
C2 to make this pole much smaller than the other poles. Therefore, the negative feedback loop operates as a single pole system and its stability is guaranteed. The same explanations hold for the lower side where a compensation capacitor can be added to the drain of M
C4 to make the related pole the dominant pole of the lower side.
Starting from
Figure 2, we can isolate a conventional Class AB voltage buffer [
30] as shown in
Figure 4. In the positive cycle, the input voltage is transferred to the output through M
11–M
12 transistors. It follows that in this case we have:
From Equation (9), to have V
out = V
in, gate-source voltages of M
11 and M
12 must be kept equal. However, the drain-source current of M
11 is a constant value of I
B8, while that of M
12 is not constant and is equal to I
L. In terms of drain currents, Equation (9) can be expressed as:
As I
dsM12 = I
L + I
B8, by assuming I
B8 << I
L, and V
thM11 = V
thM12, Equation (10) is:
Inserting I
L = V
out/R
L, Equation (11) becomes:
One effective method to reduce the non-linearity of V
out is to keep the third term in Equations (11) and (12) as small as possible. This can be realized by minimizing the variation in the drain current of M
12. Another disadvantage of the conventional class AB voltage buffer of
Figure 4 is its high output impedance, which is equal to:
Figure 5 shows the modified version of the conventional class AB voltage buffer of
Figure 4 where the current splitting method reported in [
29] is used to keep the Ids
M12 variation as small as possible. In the upper side, the current mirror made of M
15–M
16 is added and the drain of M
16 is connected to the input node. By setting the aspect ratio of M
16 n times larger than that of M
15, we have:
By assuming a positive cycle of the input signal, for I
L we have:
Inserting Equation (14) into Equation (15) we have:
Assuming Vth
M11 = Vth
M12, from Equations (16) and (10), we have:
Comparing Equations (12) and (17), we show that by applying the current splitting method, the third term in Equation (12), which is the main cause of non-linearity is reduced by a factor of (1 + n). Similarly, the current mirror made of M
17–M
18 has been added to the lower side to reduce current variation in M
14 for the negative cycle of the input voltage. Another advantage of applying the current splitting method is to reduce the output impedance by (1 + n) times:
In
Figure 5, in the upper half, there is only one dominant pole at the gate of M
15. In the lower half, the dominant pole is at the gate of M
17. In fact, the aspect ratio of M
18 is much larger (in this design 200 times) than the aspect ratio of M
14 and M
17. Therefore, at the gate of M
18, the value of parasitic capacitance is large, which makes the dominant pole of the lower half. Similarly, the large capacitance at the gate of M
16 makes the dominant pole of the upper half. Consequently, the negative feedback loops at the voltage buffer section operate as a single pole system, which is naturally stable, and no compensation capacitor is required.
The complete implementation of the high drive class AB VCII realization is shown in
Figure 6, which is constructed by series connection of the CB of
Figure 3 and the VB of
Figure 5.
3. Simulation Results
The proposed circuit of
Figure 6 was simulated in TSMC 0.18 μm CMOS technology [
32] with a supply voltage of ±0.9 V. Transistor aspect ratios are reported in
Table 1. The values of m and n were set to 10 and 200. For frequency performance analysis, a load of 50 Ω and 5 pF was connected to the X node. The Z node was connected to 10 kΩ. For frequency stability, 5 pF compensation capacitors were added between the drain and gate of M
C2 and M
C4. All bias current sources (I
B1 = 1.5 μA, I
B2,3,4,5,6,7 = 10 μA, I
B8 = 0.5 μA) were implemented with simple current mirrors with an aspect ratio of 9 µm/0.9 µm. As the current splitting section in the voltage buffer has a gain of 200, we selected a low bias current of 0.5 µA for this section. The current splitting section in the current buffer has a gain of 10 and the output branch transistors gain is 11, so we selected 1.5 μA for this section. This was done to reduce overall power consumption. The λ effect cancelling circuit operates in class A, so we selected 10 µA for this section.
The variation of V
X by I
X in the DC domain was also examined, which is shown in
Figure 7. From
Figure 7, the value of r
X is about 120.6 kΩ. The frequency performance of r
Y, r
X and r
Z are also shown in
Figure 8. The calculated and simulated values of the transistors’ small signal parameters are reported in
Table 2. Using these values, the simulation and calculation results for r
x, r
Y and r
Z are summarized in
Table 3. As it is seen, there is good agreement between the simulation and calculation results for r
x and r
Z. The achieved result for r
X is 120 kΩ for simulation and 158 kΩ for calculation. The reason for this difference is that in calculations it is assumed that the open loop gain of the λ effect cancelling circuit is assumed as infinite. However, in practice, due to its limited gain, the value of impedance at the X port is lower than expected.
To examine the frequency performance of the circuit, the X port was connected to 50 Ω resistors and a 5 pF capacitor, while the Z port was connected to a 10 kΩ resistor. The frequency performance of β and α is shown in
Figure 9. For β, the DC value and −3dB frequency were 0.993 and 11 MHz, respectively. For α, the DC value and −3 dB frequency are 0.953 and 50 MHz, respectively. The circuit power consumption was 393 μW. The DC transfer characteristic between I
Y and I
X is shown in
Figure 10. As it is seen, there is good linearity for current transfer when I
Y is varied from −10 mA to +10 mA. The linearity of voltage transfer between X and Z nodes was investigated by applying a DC voltage to the X node and connecting the Z node to a load of 10 kΩ. To compare with theory, Equation (17) was also calculated. The results of the calculations, which are shown in
Figure 11, show good agreement between theory and simulations. The limit of voltage range in node X is −0.4 V to +0.4 V.
Figure 10 shows a good linearity between X and Z nodes in the range of −0.4 V to +0.4 V. The maximum error was −70 mV, which occurred at 0.4 V.
Figure 12 shows the THD of I
X for various amplitudes of input current at 1 kHz and 1 MHz frequencies. Favorably, the value of THD for I
X did not exceed 2.4%. In fact, by increasing the frequency, the gain of the λ effect cancelling circuit reduces, so its effect starts decreasing at higher frequencies. As it is shown, the value of THD at 1 kHz is larger than 1 MHz.
Figure 13 shows the resulting THD of V
z for different amplitudes of input voltage. In this case, THD remained below 3.9%. Monte-Carlo simulations in 100 runs for mismatch of 3% between V
th and t
ox of all transistors are reported in
Table 4. In addition, corner case simulation results are summarized in
Table 5.
The application of the proposed VCII as a transimpedance amplifier [
17] is examined in
Figure 14a, while the X port is connected to a load of 100 Ω. The frequency performance is shown in
Figure 14b, which shows a gain of 16.5 dB and BW of 1 MHz.
Comparison summary of the proposed VCII and other works is shown in
Table 6. As it is seen, the proposed circuit provides the largest current drive capability ever reported. For a 29 µA bias current at M
6 and M
8, the circuit can provide ±10 mA to the X port, which is 345 times larger than the used bias current.