You are on page 1of 9

Design of four-quadrant Bipolar Multiplier

Manraj Singh Gujral, M.Sc.


School of Electronics & Computer Science, University of Southampton
SO17 1BJ, UK
msg1g10@ecs.soton.ac.uk

Abstract— This document investigates the design of Gilbert Cell.


The process includes explanation of basic concepts like Long Tail
Pair and using it to implement the Analog multiplier circuit. The
software used for simulation is LT Spice IV version 4.08u with
preloaded cell library. The final objective is to design a
multiplier giving +/1 10 V output for +/- 5V input
Keywords— gilbert cell, multiplier, long tail pair (LTP), design
flow
INTRODUCTION
In analog signal processing a need often arises for a circuit (c)
that takes two inputs and produces an output proportional to
Fig. 2 Types of Multiplier (a) Single Quadrant, (b) Double Quadrant (c) Four
their products. Such circuits are now known as the multipliers. Quadrant
A multiplier is a 2-input device which performs linear product
of two signals Vx and Vy generating an output of Vo=K.Vx.Vy. From a mathematical point of view a multiplier is a “four
From a block diagram perspective a Multiplier can be shown quadrant” operator. But in physical hardware in electronics
as in figure 1. implementing a first and second quadrant multiplier is
relatively much simpler than a four-quadrant multiplier. Also,
a lot of applications do not require a four quadrant multiplier
therefore the other types of multipliers still exist. There are
different types of multipliers being used in the industry. As an
example, AD539 is a two quadrant multiplier with two bipolar
inputs Vx1 and Vx2 and a common unipolar input Y. The
output is limited by lower bandwidth. Schematic of an AD539
Fig. 1 Block diagram of a Multiplier [1]
is shown in figure 3

Multipliers, figure 2, can be categorized as


a. Single quadrant, when both v1and v2 are positive.
b. Two quadrant, when v1 is positive and v2 is bipolar.
c. Four quadrant, when both v1 and v2 are bipolar.

Fig. 3 AD539 Block Diagram [1]

Another basic type of multipliers is the logarithmic types.


They work on the principle that the multiplication of two
numbers is addition in a logarithmic form, and the output is
(a) (b) passed though an anti-log to convert back to the same units as
the input. Figure 4 explains the basic functionality of a
logarithmic multiplier. [1]

1
therefore, ∆ =  − 
 
∆ =  (
− ) (3)
Fig. 4 Block Diagram for Logarithmic Multiplier [1]    

This is however a single quadrant multiplier with low As,


bandwidth. Therefore a much better approach, Gilbert Cell, is 1 1  /  /
− = −
1 +   1 +    / +  /  / +  /
used to explain the multiplier operation.
Before moving further we should understand the difference
between a modulator and a multiplier. A modulator is a circuit
 / −  /
which can have a number of inputs and one input can be used = tanh (/2)
to modify or control the signal flow from another input.  / +  /

So, Eqn. (3) can be re-written as,


!
∆ =  tanh ( ) (4)[2]
!

Fig. 5 Block Diagram for a Modulator


Assuming the Applied voltage, Vid<< VT (for small signal
A multiplier is a two input device which is independent of analysis) we can write
the input signals and performs a multiplication by a factor, k, ∆ =  ("#$ /2"% ) (5)
dependent on the multiplier parameter. Therefore in order to control the Ic we can control the IEE,
making IEE the bias current for the Long Tail pair. Here we see
that If we can control the IEE externally, we can alter the
biasing of the LTP. So we add extra circuitry to make IEE
proportional to an external voltage source, as shown in figure
8

Fig. 6 Block Diagram for a Multiplier

Therefore, it can be said that a modulator is a special case


of a multiplier.

LONG TAIL PAIR [2]


An important concept used in any multiplier is the use of a
Long-Tail Pair, also known as the emitter-coupled pair circuit.

Fig. 8 Long Tail Pair with controlling voltage

 ≅ '( ("# − ")((*) ) (6)


Fig. 7 Long Tail Pair or Common Emitter Pair
So now, the differential output current at Eqn. (4) of the
Figure 7 shows a basic construction of a Long Tail Pair. LTP can be re-written as,
The outputs current, Ic1 and Ic2 are related to the differential
"
voltage by: + ≅ '( ("# − ")((*) )tanh (2",- )

 =  (1) .
(7)[2]


 


 = (2)
( )
 

2
As we can observe from the waveform, figure 12, that this
particular multiplier is basically a two-quadrant multiplier. All
the output values for the multiplier remain at 0V for any
voltages of V5 < 0.

Fig. 9 Long Tail Pair circuit realization

The output, Eqn. (7), is a multiplier assuming that Vid is a


small signal and Vi2> VBE. In circuit realization in figure 9,
V5 is sweep from 1 to 5. Instead of using a pure differential Fig. 11 LTP circuit with voltage sweep for V5 from -5 to +5
voltage Vid, we use two voltages V4 and V3 w.r.t ground and
fix on of them to 1V and sweep the other for simulation.

Fig. 10 Long Tail DC sweep waveform Fig. 12 Waveform analysis for voltage sweep for V5 from -5 to +5

As seen from the wave in figure 10 the voltage (V5) has to The Voltage V5 has to be positive at all times since the IEE ,
be positive (1V to 5V). As from the waveform we can see the seen in figure 8, has to flow in order for the LTP to function.
central point when input voltage is 1 V, i.e. when V1 = V2 the This causes further restrictions on its usage. Therefore we will
differential voltage output is zero. Also the curve stays linear use this principle of a cross coupled Long Tailed Pair to
for very small range of differential input voltage (v4-v3) in the modify the current circuit and use it in all four quadrants.
range of +/- 0.6V. And beyond that the output voltage is
ANALYSIS OF GILBERT MULTIPLIER
saturated at fixed level.
Analysing a random output wave, say at V5=4, the output Barry Gilbert in 1967 had invented an Analog multiplier,
differential voltage / 6.1V and is constant for any change in which is now known as the Gilbert Cell, or the Gilbert mixer
the input. In the liner range, when V3 in in the range 0.96V to [3]. A basic structure of a Gilbert Cell is two Long Tailed
1.04V, we can assume the value of K is constant we can Pair connected in series with one half of other long tailed pair
therefore generate a multiplier with the equation as mentioned as shown in the figure 13
in the Introduction
Vo=K.Vx.Vy
We can adjust the factor K by varying the tail current or the
resistances values R1 and R2. In a practical circuit, however,
it is prudent to have a voltage controlled reference that a
variable resistor which we will use later in this report.
Now we perform a large signal analysis on the same circuit
and a voltage sweep for both negative and positive values of
input voltage V5. From -5 to +5 in steps of 1. The circuit is
shown in figure 10

3
,
IL1 IL2 ,4 = ,
;1 +  (< ⁄<% ) >;1 +  (<⁄<%) >

,
I c3 I c4 I c5 I c6

,6 = ,
+ Q1 Q2 Q3 Q4
;1 +  (<⁄<%) >;1 +  (<⁄<%) >

,
v1
_
,7 =
I c1 I c2 ;1 +  (<⁄<%) >;1 +  (<⁄<%) >
(11)
+ Q5 Q6 Assuming the Change in collector currents as
v2
_ ∆ = ? − ? = (,0 + ,6 ) − (,4 + ,7 )
IEE
= (,0 − ,7 ) + (,6 + ,4 )
and substituting the values of individual collector currents
from Eqn. (11) we get,
1 1 1
Fig. 13 The basic structure of the Gilbert Cell ∆ =  @A − BC @A
1+ < ⁄<% 1+ < ⁄<% 1 +  <⁄<%
To understand the basic operation, consider V2 as reference 1
− BC
1 +  <⁄<%
and V1 as input. If V1 is a positive voltage and large enough
that transistor Q1 and Q4 are powered on and Q2 and Q3 are
<2 <9
therefore powered off. Therefore as is the case in a standard =D ∆ =  tanh ( )tanh ( ) (12)[5]
differential amplifier, < <
(a) Q5 is connected to IL1 load line, and
(b) Q6 is connected to IL2 load line. This Eqn. (12) forms the equation of Gilbert Cell where the
Whereas, if the input voltage is a negative voltage large output is the product of hyperbolic tangents of the input
enough to power on Q2 & Q3, and hence, power off Q1 and voltages.
Q4 then we have We use the same concept in out circuit realization as shown
(c) Q5 is connected to IL2 load line , and in figure 14. Instead of the Current source IEE, we employ a
(d) Q6 is connected to IL1 load line. current mirror controlled by a voltage reference. Note that Q5
Which is also the standard differential amplifier, but with and Q6 now act as separate arms for the current to flow for V3
negative polarity. [4] to be negative and positive respectively. Transistor Q5 , in
Theoretically it should work for both the negative and figure 14, is responsible for the flow of current labelled as Ic1
positive values of the input and output giving us the four in figure 13 when the V3 is positive. Transistor Q6, in figure
quadrant operation. We will now look at the mathematical 14, is responsible for flow of current Ic2 when V3 becomes
derivation of the output and later simulate and see the results negative. Actual biasing of the transistors needs to be
to prove it works. appropriate since a negative differential voltage effectively
As in Eqn. (4), here we have two voltage controlling the means that one of the transistors, say Q5, is active during the
output namely v1 and v2. positive voltage cycle and therefore Q6 in negative voltage
cycle since it sees a positive potential w.r.t. Q5.
In the figure 13 , we have
, ,
,0 = 12 , ,4 = 12 (8)
1 +  (1) 1 +  (1)

and,
, ,
,6 = 12 , ,7 = 12 (9)
( )
1+ 1 1 +  (1)

And the second LTP currents as,

, ,
, = 19 , , = 19 (10)
( )
1+ 1 1 +  (1)

Substituting Eqn. (10) in (8) & (9) we find that


,
,0 = ,
;1 +  (<⁄<%) >;1 +  (<⁄<%) >

Fig. 14 The Gilbert Cell – two cross coupled LTP

4
(b)
Fig. 16 Two cross coupled LTP waveform, with Ic(Q7) = 6.25 mA. (a) Circuit
with V4=4V, (b) Resulting waveform and Ic(Q7

From figure 16 we can observe the output has changed


from a range of +/- 2.4 V to +/- 6 V. Hence, our equation,
Eqn.(14) holds true.
This is just an operational schematic intended to simplify
and explain the Gilbert Cell operation. As we move to a more
practical aspect, we would need to provide an actual
differential voltage across the cross coupled LTP instead of
two voltage sources, V4 and V6.
The small signal analysis of the same circuit as shown in
Fig. 15 Two cross coupled LTP waveform, with Ic(Q7) = 2.35mA. figure 17
Observing from the waveforms in figure 15, we now have
operations in all the four quadrants, i.e., the negative and
positive values of the input voltages.
Please note that we have maintained the reference V6 fixed
at 1V and therefore the changeover occurs at 1V because V4-
V6 = 0 at 1 V.
In figure 14 & 15 the current flowing current mirror arm
into Transistor Q7 is
Ic(Q7) = 2.35mA . (13)

As per Eqn.(12), we see that the output voltage is basically


dependent on three factors, Fig. 17 Small Signal analysis of LTP waveform, with Ic(Q7) = 6.25 mA
∆ ∝  . tanh( G ). tanh (G ) (14) V3 from -0.1V to 0.1V in steps of 0.01V
For the sake of validating our equation we will change the
reference voltage V5 to control IEE and observe the output Small signal analysis shows that the waveform follows the
Tanh graph and therefore the liner range is very small.

We have now established the basic working of a Gilbert


circuit. From figure 17 it can be observed that it shows the
features of a multiplier circuit. With small signal input voltage
of we get a linear characteristic output. In order to move to a
more practical circuit we need to properly bias the inputs ,
increase the linearity of the output and have some
controllability. Therefore, we start by replacing the Voltage
source V4 with an inverse hyperbolic tangent differential
voltage source. Details of the effect of introducing voltage-to-
current converter are given in figure 21. As of now, we were
using V4 and V6, see figure 16 (a), to supply the input voltage.
Since V6 is always fixed at 1V the output was seen across the
1V mark which was acting as the reference point of the circuit,
figure 16(b). As we now move to practical multiplier like the
one shown in figure 18 which is a very basic from multiplier,
(a) we need only VX1-VX2 as a differential voltage source.

5
Fig. 18 Low cost analog multiplier AD633, 8-Lead Plastic DIP (N) Package

Also, instead of a differential output, we could employ the Consider a simple Common Emitter configuration:
circuit for a single ended output using current mirrors. Vcc
Therefore, implementing a differential voltage input and using
Rc
emitter degeneration in place of Q7 & Q8 we get the
following characteristics as shown in figure 19. Ic
Rb
Vb
Vb
Re

Fig. 20 Basic CE Configuration of an npn transistor [4]

For dc Bias , the KVL loop can be written as:


"H = H IH + "H +  I
Assuming  = JH , if Re=0 then the KVL equation
becomes,
(a) IH
"H / H IH + "H =  + "H
J
i.e., the Bias current Ic depends on the transistor β.

Whereas assuming  = JH &  /  if ReK0 then


the KVL equation becomes,
"H / H IH + "H +  I = "H +  I
Since IbRb << IcRe . Therefore Vb controls Ic without
depending on the transistor β making the biasing
more stable. [4]
Also, it can be shown that the Output gain of the
common emitter amplifier (small signal)becomes :
(b) "(MN I
L< / =−
Fig. 19 Gilbert Cell with differential Inputs and single ended output (a) "#* I
Circuit diagram, (b) Resulting output waveform
And this depends only on the external resistors and
The output characteristics have been modified and the independent of transistor β. [4]
observations are: We will try and figure out the output voltage relation to these
(a) Input voltage is now passed to the gilbert Cell as a circuit elements by performing certain tests.
differential voltage.
(b) The output is taken across the Resistor R5 as a single Differential-to-single-ended
ended output. The circuit implementation is nothing converter
but two current mirrors connected to the ∆ arm of
Vout
Vcc IL1 IL2
the previously used circuit, figure 16(a). The current
mirrors both try to maintain the same current in its Ic3 Ic4 Ic5 Ic6
arm and the differential current is passed through the
resistor R5 resulting as the output. Q1 Q2 Q3 Q4
(c) The output range of the voltage is now < +/- 2V for
and input voltage = +/-5 V.
(d) The Output voltage is now dependent on Resistors, + Differential voltage-to-
v1
_ current converter
R2 & R5, the input and output resistors resp. as well Ic1 Ic2
as R4 the emitter degeneration resistor. + Differential
Q4 voltage toQ5
current
v2 converter
(e) The Linear range of the circuit is drastically _
improved from figure 12, the range is now valid for
the input voltage range of +/-5 V. Fig. 21 Basic Gilbert Cell Schematic [6]
(f) Addition of a degeneration emitter resistor not only
In figure 19, all we have done is provided the Basic gilbert
improves the liner range, but also makes the system
cell with proper differential voltages and single ended output,
more stable since the Ic is no longer dependent on the
figure 20.
temperature varying parameter of the BJT, β.

6
Z2 !2
(a) The single ended output is nothing but two ∆"  2"% tanh Y [ (20)
current mirrors trying to balance each other and P
Substituting Eqn. 20 in Eqn.12 we get ,
G G
the differential current flowing through the Load
∆   tanh A B tanh A B D
2G% 2G%
resistor. Addition of a differential-to-single-
ended converter adds a parameter in the circuit
Z ! Z !
as ∆   Y 2 2 [ Y 9 9 [ (21)[6]
P OP
"(MN = ∆ . ' (15)[6] This makes it linear to the Input voltage rather than
hyperbolic tangential.
Where K is dependent on the load resistance and the
voltage controlled current source in the differential-to-single Our specification is to meet a specification of +/- 10 V
ended converter. Output for an Input range of +/-5V with flexibility of
(b) Since differential voltage-to-current converter controlling the tail current in the cross connected LTPs.
with input V2 is responsible for the tail currents Using Eqn.15, 16 & 17 to vary the circuit parameters, such
in the cross coupled LTP, it has an important as the input and output resistances; the tail current settings;
role in output where and the differential voltage V3 controlling the tail currents of
!
"(MN ∝ 9 (16) cross coupled LTP in figure 18 is modified to get the desired
OP
The Ieo current in Eqn.(16) can be seen in our simulated output.
circuit, figure 19(a), as the degenerated emitter current source
as, I3 & I4
(c) Similarly, the differential voltage-to-current
converter with input V2 is the main input signal
which controls the LTP pair transistors.
Therefore addition of this converter has an effect
on the output as
!2
"(MN ∝ (17)
 Q RRPR STRROUV
The Imirror current in Eqn.(16) can be seen in our simulated
circuit, figure 19(a), as the degenerated current source as, I1 &
I2. Practically the purpose of this differential voltage-to-
current converter is to introduce an inverse hyperbolic (a)
Tangent voltage to cancel the effect of the Gilbert Cell’s
equation given in Eqn. (12).

Vcc

∆V

I1 I2

+ Differential voltage-to-
v1 (b)
_ current converter
Fig. 23 Analysing the output by varying circuit parameter R2
Fig. 22 Inverse hyperbolic tangent circuit [6]
(a) Replacing the resistance R2 from 5k to 10k (b) Resulting output
The differential voltage source produces a current in waveform V(output) vs. Input voltage, V4
addition to the I1 and I2 shown in the figure 21. The currents
Figure 22 shows the response of the circuit when we vary
are hence given by,
the resistance R2 from its set value of the circuit (shown in
 = ( + ' " , and  = ( − ' " (18)
after Figure 25). The output swing is restricted to +/- 5V from
Here Io is the current when V1= 0V, and K1 is the factor
+/- 10V proving the output is inversely proportional to the
introduced by the voltage-to-converter.
Therefore, the differential voltage ∆" is given by,
input degenerated resistor.
(  ' " (  ' "
∆"  "% ln A B  "% ln A B
X X
 Z !
 "% ln Y P 2 2 [ (19)
P Z2 !2
 
Using, tanh 
  ln Y [
 
We have,

7
(b)
Fig. 25 Analysing the output by varying circuit parameter R5
(a) (a) Replacing the resistance R5 from 5k to 10k (b) Resulting output
waveform V(output) vs. Input voltage, V4

Figure 24 shows the response of the circuit when we vary


the resistance R5 from its set value of the circuit (shown in
after Figure 25). The output swing is doubled to +/- 20V from
+/- 10V proving the output is directly proportional to the
differential-to-single-ended converter factor given by Eqn. (16)

(b)
Fig. 24 Analysing the output by varying circuit parameter R4
(a) Replacing the resistance R4 from 1.8 to 3.65k (b) Resulting output
waveform V(output) vs. Input voltage, V4

Figure 23 shows the response of the circuit when we vary


the resistance R4 from its set value of the circuit (shown in
after Figure 25). The output swing is restricted to +/- 5V from
+/- 10V proving the output is inversely proportional to the
Emitter degenerated resistor (validating Eqn. (16)).
(a)

(b)
Fig. 26 Gilbert Cell with adjusted circuit parameters. (a) Circuit diagram, (b)
Resulting output waveform V(output) vs. Input voltage, V4
(a) The final required output is obtained in Figure 25, for input
voltage swing of +/-5V the output ranges from +/- 10V. the
circuit also has a voltage controlled Tail current controller , as
V3, in the circuit which can adjust the multiplication factor.

Gilbert Cell described so far has been BJT based, the same
model can be developed for a MOS based Cell as well.

8
voltages as was the case with bipolar based multiplier, rather
it is a linear and directly proportional to the inputs

LEARNING OUTCOMES & CONCLUSIONS


Gilbert Cell can be used for wide range of applications in
electronics & communication. While working on this report a
large number of its applications were found working on the
principle of this cell, such as:
1. Frequency modulator where the V1 can be a modulating
small signal and the Carrier large signal can be supplied in V2
to get modulated output.
2. Phase detector, where the product of both the inputs
(both large signals) gives the out of phase voltage at the
output.
Use of current mirrors, alone and in conjunction with the
Long Tail Pair has many practical uses and is a very important
Fig. 27 MOS based Gilbert Cell [7] tool in electronic circuit design today. The individual concepts
Here Iout = I1-I2 for LTP and Current mirrors are not covered in detail in this
Where currents through the nMOSes are given by, report but their functionality is quite apparent and similar to
Z2  Z9 
\  ]"^X  "% _ , \  ]"^X  "% _ , the Gilbert Cell, they can be used in variety of modules.
 
Z`  Za 
\0  ]"^X0  "% _ , \4  ]"^X4  "% _ (22)
 

It is essential that the transistors are matched (common REFERENCES


centroid techniques) i.e., [1] Analog Devices - MT-079 TUTORIAL Analog Multipliers.
d [2] Gray, P.R. & Meyer. R.G. (Dublin Institute of Technology), Analog
'  '  '0  '4  b* c(
e
Multipliers, Analysis and Design of Analog Integrated Circuits (3rd
Edition), Wiley (1992) pp. 667-681.
If we assume the LTP functioning of the circuit [3] “The Gears of Genius” , SSCS IEEE Solid-State Circuits Society News;
2007 Volume 12, Number 4
[4] Prof. Keith W. Whites, Electronics II-Wireless Communication
Electronics (EE 322), Lecture 27 & 19 ,Dept. of Electrical & Computer
engineering, South Dakota School of Mines & Technology
[5] Analog Circuits and Systems Design (ECE6414) - Multipliers and
PLLs , ECE , Georgia Tech University,
[6] Dr. Mohamed Abd El Ghany, Communication Systems (COMM 704).
Department of Electronics and Electrical Engineering, German
University in Cairo.
[7] Associate Prof. Dr. Soilam Mahmoud, Microelectronics Lecture 2 & 3,
winter 2006, Electronics and Electrical Department , German
University Cairo

Fig. 28 MOS based Gilbert Cell [7] BIBLIOGRAPHY


[1] Gunhee Han and Edgar S´anchez-Sinencio, CMOS Transconductance
The output current can be shown to be
Multipliers: A Tutorial, Fellow, IEEE, IEEE transactions on Circuits
!29
?%f(MN  \  \  gXX '" hi1  k
and systems –II: Analog and digital Signal Processing, Vol.45, No.12,
4jj⁄Z December 1998.
[2] Barrie Gilbert, Current Mode, Voltage Mode, or Free Mode?, Analog
(23) Integrated Circuits and Signal Processing, 38, 83–101, 2004, Kluwer
Academic Publishers. Manufactured in The Netherlands
And similar to Eqn. (12) derivation the Gilbert Cell Output [3] Datasheet, Rev. D, AD633, Low cost Analog Multiplier, Analog
current can be written as Devices.

(MN      \  \   \0  \4  (24)


Z
(MN ≅ " . "  (25)[7]
√

One important factor to note here is that the Output is no


longer a the product of hyperbolic tangents of the input

You might also like